Midjourney is a cutting-edge AI-powered service, created by an autonomous research laboratory that shares its name, which specializes in transforming textual input into visually stunning images. By enabling users to produce images that correspond to their written descriptions, this innovative technology caters to an extensive array of artistic expressions, ranging from lifelike representations to more abstract and imaginative styles.
How does Midjourney work?
Midjourney operates through a sophisticated artificial intelligence process that leverages deep learning algorithms to analyze and interpret textual input provided by users. The AI system is trained on vast datasets consisting of text-image pairs, which helps it understand and establish relationships between words and visual elements. As a result, it can accurately generate images based on the descriptions it receives.
When a user submits a textual description to the Midjourney platform, the AI system parses the text and identifies key elements, such as objects, colors, and contextual information. It then uses this understanding to generate an initial image that reflects the user’s description. The AI continually refines and adjusts the image based on its deep learning capabilities and the context it has learned from its training data.
Furthermore, the Midjourney AI also takes into account various artistic styles and patterns, allowing it to create images that range from realistic to abstract. This flexibility empowers users to explore their creativity and produce unique, customized visuals that align with their vision.
Can you use Midjourney for free?
Midjourney recently suspended their free tier because of abuse.
Still, we think Midjourney is a great value at the lowest cost for businesses or artists with even a little bit of image needs.
How much does Midjourney cost?
Midjourney currently has 3 plans — $10/m, $30/m, and $60/m with yearly subscription discounts. Check this out: https://docs.midjourney.com/docs/plans
How do you use Midjourney AI? Guide to Using Midjourney
To begin utilizing Midjourney, follow these simple steps:
Firstly, ensure you possess a Discord login, as Midjourney functions exclusively on this platform. If you haven’t signed up yet, it’s free and easy to do.
Next, navigate to the Midjourney website and opt for “Join the beta.” This action will direct you to a Discord invite.
Accept the Discord invite to gain access to Midjourney.
Upon launching the Discord app, locate the Midjourney icon (resembling a ship) in the left-hand menu.
Find the Newcomer rooms within the Midjourney channels. Select any available room, such as “newbies-108,” to get started.
Before diving into creating AI-generated artwork, be mindful of your limited prompt options during the free trial. With approximately 25 free images at your disposal, it’s wise to carefully plan your creations. For a list of useful tips, type “/help.”
When you’re prepared, type “/imagine” in the Discord chat for your designated room. Enter a detailed image description to aid the AI in generating the best possible outcome. Adhere to the terms of conduct and press Enter to submit your prompt.
Allow Midjourney a moment to produce multiple image variations. Underneath the images, you’ll find U and V buttons labeled 1 through 4, which correspond to the generated images. To upscale or generate a new image based on your selection, click the appropriate button. Keep in mind that each choice consumes a portion of your free prompts.
If you decide to upscale an image, you’ll encounter additional options, such as making variations, upscaling to the maximum before downloading, or opting for a light upscale redo. To save and download the image, send it to yourself by selecting the Envelope emoji.
For avid Midjourney users, consider subscribing by typing “/subscribe” in any bot channel within the Discord server. This command generates a payment link for a subscription. To further enhance your experience, consult the manual for an expanded list of commands and image creation tips.
How can I access Midjourney?
Login to midjourney.com to sign up for your account and then you will need to follow their instructions to access the Midjourney bot via Discord.
Can you use Midjourney for NSFW?
Not really because they have NSFW blocks on the prompts. Sometimes you can get something a bit R rated, but there’s no direct way to do it.
Text-to-image AI models are a type of generative model that can create images based on textual descriptions. These models typically use deep learning techniques, such as neural networks, to learn the associations between text and visual content. They are trained on large datasets of text-image pairs, where they learn the patterns and structures that correspond to the textual descriptions.
Here is a high-level overview of how text-to-image AI models work:
Tokenization and Embedding:
The input text is first tokenized into words or smaller units, such as subwords or characters. These tokens are then converted into numerical representations, called embeddings, which can be fed into the neural network.
Encoder:
The text embeddings are passed through an encoder, which is typically a type of neural network (such as a Transformer or a recurrent neural network). The encoder processes the input text and generates a high-level, context-aware representation of the text, which captures the essential features and semantics of the input.
Decoder:
The decoder is another neural network that takes the high-level text representation from the encoder and generates an image. It does this by producing a series of feature maps, which are then used to build the final image. The decoder may use architectures such as convolutional neural networks (CNNs), generative adversarial networks (GANs), or variational autoencoders (VAEs).
Loss Function and Training:
The generated image is compared to the ground truth image (i.e., the correct image associated with the input text) using a loss function. The loss function quantifies the difference between the generated image and the ground truth image. During training, the model learns to minimize this loss by adjusting its parameters, which helps improve the quality of the generated images.
Sampling and Post-processing:
Once the model is trained, it can generate images for new text inputs. The output image is usually obtained by sampling from the probability distribution of pixel values produced by the decoder. Some post-processing techniques, such as upsampling, denoising, or colorization, may also be applied to refine the generated image.
What are the Most Popular Text to Image AI Models?
It is the opinion of 42 Robots (and many others), that in May of 2023, for quite some time Midjourney is the best overall AI Image Generator.
What are Common Problems with Image AI Models?
Text to Image AI models are amazing, but most come with some common problems:
1) Two Heads
Some of the image AI models produce people with 2 heads. Obviously, this is a significant issue. Fortunately, it appears less and less common.
2) Not Showing Full Body
AI models tend to want to show just the upper body, so you often have to be very specific to get a full body image.
3) Garbled Faces
Faces often come out clearly not right. This is improving rapidly. Midjourney does a great job of creating realistic faces.
4) Extra or missing fingers or hands
Hands are very difficult for AI to get consistently right. Sometimes whole hands are missing and sometimes the AI model has created a 6 fingered person. Also, getting better rapidly.
What are Good Use Cases for Text to Image AI Models?
Text-to-image AI models have many potential applications, including art generation, advertising, content creation, and data augmentation. However, it is important to note that the quality of the generated images depends on the quality of the training data and the architecture of the model.
Text-to-image AI models have a wide range of applications across various domains. Some good use cases include:
Art and Design:
Artists and designers can use these models to generate creative and unique visual content based on textual descriptions, which can serve as a starting point for their projects or inspire new ideas.
Advertising and Marketing:
Text-to-image models can be used to create tailored visuals for ad campaigns based on specific target audience descriptions or product features. This can help streamline the content creation process and enhance the effectiveness of marketing campaigns.
Virtual Reality and Gaming:
In virtual environments and video games, text-to-image models can generate realistic images, scenes, or objects based on user input or game narratives. This can enhance the immersive experience and facilitate the development of dynamic, user-driven content.
Data Augmentation:
Text-to-image AI models can generate additional training data for other machine learning models, particularly when there is limited visual data available. By creating new images based on text descriptions, these models can help improve the performance of computer vision models or other AI systems.
Concept Visualization:
Researchers, scientists, and engineers can use text-to-image models to visualize abstract concepts or complex ideas that are difficult to represent graphically. This can aid in communication, education, and knowledge transfer.
Content Creation and Storytelling:
Text-to-image models can help create engaging multimedia content, such as illustrations for books, articles, or social media posts. They can also be used in storytelling by generating visuals based on narrative descriptions, helping to bring stories to life and enhance the reader’s experience.
Fashion and Retail:
Text-to-image AI models can generate images of clothing or accessories based on textual descriptions of styles, materials, or colors, which can be useful for fashion designers or online retailers to showcase their products.
Customized User Experience:
Text-to-image AI models can be integrated into user interfaces to generate personalized content or recommendations based on user preferences or input.
Which AI Models can do Image to Text?
Some AI models go in the reverse direction, Image to Text. In other words, they are AI models that can describe an image. GPT-4 and Midjourney both do this. There are other Image to Text AI Models.
Hallucinations in large language models refer to the generation of text or responses that are not grounded in facts, logic, or the given context. While these models, like GPT-4, have been trained on vast amounts of data, they are still imperfect and sometimes produce incorrect or nonsensical answers.
Delusion & confabulation are alternative terms for hallucinations.
Some in AI suggest that “hallucination” is not a good term because it implies more human qualities than the LLM really has.
Why does an LLM Hallucinate?
There are a few reasons why hallucinations might occur:
Incomplete or noisy training data:
Since language models are trained on data from the internet, they are exposed to a wide range of information, including misinformation and inaccuracies. This can lead to the model generating factually incorrect or nonsensical responses.
Over-optimization:
Language models try to maximize the likelihood of the next word in a sentence, given the previous words. This can sometimes lead to over-optimization, where the model prioritizes fluency and coherence over factual accuracy.
Lack of context:
Language models may not have a deep understanding of context, leading to situations where they generate text that is plausible-sounding but incorrect or unrelated to the input.
Ambiguity in user input:
If the input provided to the language model is ambiguous or unclear, the model may generate a response based on its best guess or interpretation, which can result in a hallucination.
How Can We Fix Large Language Model Hallucinations?
Researchers and developers are actively working to address these challenges and improve the reliability and accuracy of large language models. Techniques like providing more accurate training data, incorporating external knowledge bases, and developing better methods for grounding responses in context are some approaches being explored to minimize hallucinations in these models.
How Often do LLM Hallucinations Occur?
The frequency of hallucinations in large language models (LLMs) can vary depending on several factors, including the specific model, its training data, and the nature of the input provided. It is difficult to provide an exact number or percentage for how often hallucinations occur, as this can vary widely based on the context and use case.
Generally, state-of-the-art models like GPT-4 are designed to be more accurate and less prone to hallucinations than their predecessors, thanks to improvements in architecture, training techniques, and the use of larger datasets. However, even these advanced models can still produce hallucinations under certain circumstances.
Factors that can influence the frequency of hallucinations include:
Ambiguity in input:
If the input provided to the LLM is unclear or ambiguous, the model may be more likely to generate a response that is not grounded in facts or context.
Complexity of the topic:
Some subjects or questions are inherently more complex or open to interpretation, which may increase the likelihood of hallucinations.
Quality of training data:
If the model has been trained on noisy or inaccurate data, it may be more prone to hallucinations.
Model size and architecture:
Larger and more advanced models may generally produce fewer hallucinations than smaller or less sophisticated models.
Researchers and developers are constantly working on improving the performance of LLMs and reducing the occurrence of hallucinations. This is an ongoing area of research, and as models continue to improve, the frequency of hallucinations is expected to decrease over time.
Narrow AI, or Weak AI as it’s often called, are algorithms designed to carry out specific functions.
At 42 Robots, we think calling narrow AI, “weak” is incorrectly labeled as the narrow AI models are and can be incredibly powerful.
What is Strong AI?
The objective of Strong AI is to develop intelligent machines that seamlessly replicate the human intellect.
Does strong AI exist today?
It is the opinion of 42 Robots that GPT-4 is an example of “strong” AI or it is at least close enough to seriously consider it for the label.
Artificial Intelligence (AI) conjures images of robots and futuristic technology, but today’s AI is mostly about Narrow AI. With a capacity to solve complex problems and outperform humans, Weak AI’s capabilities remain restricted by its programming. Despite its seemingly limited name, Narrow AI pervades various aspects of our lives, from work to leisure. Here’s a closer look at Weak AI and eight intriguing examples.
Weak AI vs Strong AI
Weak AI and Strong AI are two types of artificial intelligence that differ in terms of their capabilities, goals, and potential applications. Here’s a detailed description of each:
Weak AI:
Definition: Weak AI, also known as Narrow AI, refers to AI systems that are designed and trained to perform specific tasks or solve particular problems without possessing general intelligence. These systems are not self-aware and do not have the ability to understand or apply their knowledge beyond the tasks they were designed for.
Capabilities: Weak AI systems excel in the tasks they are designed for, such as image recognition, natural language processing, or playing games like chess. They can process and analyze large amounts of data quickly and accurately, but their capabilities are limited to the domain they were trained on.
Learning: Weak AI systems are typically trained using supervised learning techniques, where they learn from labeled data sets. They can also employ unsupervised or reinforcement learning, but their learning is constrained within the task or problem they were designed to solve.
Examples: Some examples of Weak AI include chatbots, recommendation systems, virtual assistants like Apple’s Siri or Amazon’s Alexa, and self-driving cars.
Ethical concerns: While there are ethical concerns surrounding Weak AI, they are generally less severe than those related to Strong AI. Issues such as data privacy, algorithmic bias, and job displacement are still relevant, but Weak AI does not pose existential threats to humanity.
Strong AI:
Definition: Strong AI, also known as Artificial General Intelligence (AGI), refers to AI systems that possess human-level intelligence or beyond and can perform any intellectual task that a human being can do. These systems are self-aware, can understand context, and adapt their knowledge and learning to new domains.
Capabilities: Strong AI systems can perform a wide range of tasks, including problem-solving, reasoning, learning, planning, and creativity. They can understand natural language, perceive and interpret their environment, and interact with the world in a human-like manner.
Learning: Strong AI systems can learn and adapt autonomously, using techniques such as unsupervised learning, reinforcement learning, and transfer learning. This enables them to generalize their knowledge and apply it to new domains, problems, and tasks without being explicitly programmed to do so.
Examples: As of my knowledge cutoff in September 2021, Strong AI has not yet been achieved. However, researchers and developers are continuously working towards creating AGI, which would be a significant milestone in the field of AI.
Ethical concerns: Strong AI raises a host of ethical, philosophical, and existential concerns. These include the potential loss of control over AI systems, the impact on human dignity and employment, AI consciousness and rights, and the possibility of an intelligence explosion leading to a “singularity” event.
In summary, Weak AI refers to task-specific AI systems that excel in their designated domains but lack general intelligence, while Strong AI represents AI systems with human-level intelligence or beyond, capable of performing any intellectual task. While Weak AI is prevalent in many applications today, Strong AI remains a long-term goal for AI researchers and developers.
Here’s a quick video overview of Strong AI vs Narrow AI:
Unraveling the Enigma of Weak (Narrow) AI
At the heart of Narrow AI lie algorithms that emulate human intelligence, focusing on accomplishing particular tasks instead of exhibiting full cognitive abilities like the human brain. These algorithms are trained to sort data based on specific instructions, sticking to their pre-defined functions. For instance, a machine designed for warehouse item picking wouldn’t be able to cook burgers without reprogramming. Narrow AI models intelligent behavior for specific tasks.
Which AI is considered as weak AI? 7 Narrow AI Examples.
Today’s AI largely consists of Narrow AI, which is seen in various practical applications rather than sci-fi novel scenarios or film depictions of robots dominating the world. Here are eight fascinating examples:
Digital voice assistants (Siri, Alexa) As prime examples of Weak AI, digital voice assistants like Siri and Alexa swiftly classify data and respond to our queries, becoming indispensable in our daily lives.
Recommendation engines Narrow AI powers recommendation engines that suggest movies on Netflix or provide shopping advice on Amazon and other retail websites.
Search engines Google and other search engines are also examples of Weak AI, which swiftly classify and find answers to our queries.
Chatbots Chances are, if you’ve used a chat feature with a company, you’ve conversed with AI. These chatbots employ AI algorithms to answer common questions, freeing up humans to tackle more complex tasks.
Autonomous vehicles Weak AI steers vehicles without human drivers by executing programmed functions. The challenge lies in training AI to deal with any potential road hazards or situations.
Image and speech recognition Narrow AI significantly impacts healthcare by aiding radiologists in detecting diseases through image recognition. It’s also utilized in speech recognition and translation services like Google Translate.
Predictive maintenance and analytics Narrow AI examines historical data to predict future outcomes using data, algorithms, and machine learning. It also helps identify maintenance issues in warehouses and other heavy machinery environments before machine failures occur.
Narrow AI is the technology that surrounds us in our daily lives. Powered by algorithms, it enables machines to act, process data, and make decisions based on programming. Although its name might suggest otherwise, Weak or Narrow AI is responsible for numerous impressive feats, often outpacing human efficiency.
GPT-5 is the theoretical next generation AI Large Language Model (LLM) from OpenAI.
GPT-3.5, living inside ChatGPT’s first publicly available version, amazed the world.
And then GPT-4 came out and took things to a whole other level.
Immediately, speculation about GPT-5 began.
When is GPT-5 coming out?
In April of 2023, Sam Altman, implied that GPT-5 is not close to being released. Based upon many factors, we think it is likely to come out in late 2024 or in 2025.
Will GPT-5 come out at all?
Sam Altman also stated that the “Age of Giant Models is Over” when talking about the diminishing returns of making the LLM’s exponentially bigger.
Therefore, it is possible that there will never be a GPT-5. Perhaps it will be a new acronym version 1.
More details on what Altman said:
Will GPT-5 be free?
Considering when GPT-4 came out, it was not available for free (without access to temporary backdoors), we expect GPT-5 to be paid as well.
How powerful will GPT 5 be?
Oof! GPT-4 is so incredibly impressive that it is hard to fathom how good GPT-5 is at the time of writing this (May 2023).
If we were to take an educated guess, we would suggest that many of the issues that plague GPT-4 will be dramatically improved, if not completely solved.
For example, perhaps they will have hallucinations (perhaps GPT-4’s biggest issue) fixed.
Will GPT-5 be an AGI (Artificial General Intelligence)?
Watch this excellent video about GPT-5, AGI, and related topics:
DALL-E is a version of GPT-3, an artificial intelligence model developed by OpenAI, specifically designed to generate images from textual descriptions. It was trained on a dataset containing pairs of text and image data, which allows it to generate unique, creative images from any text prompt it is given.
For instance, if you were to give DALL-E a prompt like “an armchair in the shape of an avocado,” it would generate an image that fits that description. Its ability to synthesize novel images from unusual and even nonsensical prompts shows its ability to understand and combine different concepts in a way that was not previously seen in AI models. As of my knowledge cutoff in September 2021, DALL-E’s capabilities are quite impressive, but it’s important to note that the technology is still being developed and refined.
One important thing to mention is that while DALL-E can create amazing and unexpected images, it doesn’t “understand” the concepts in the way humans do. It doesn’t have a consciousness or a subjective understanding of the world; it’s simply generating output based on patterns it learned during its training.
What is DALL-E 2?
DALL-E 2 is an advanced AI image generator that can create images based on textual prompts, although the results can sometimes be unpredictable. Despite its impressive capabilities, it is not seen as a replacement for photographers, Photoshop, or other artists. The AI tool, developed by OpenAI, is available for free use and is considered to be an excellent resource for generating art inspiration.
DALL-E vs DALL-E 2
DALL-E 2 is similar to DALL-E, but, generally with higher quality.
Speedy Symbiosis of Visuals & TextDALL-E 1, the artist of the AI world, crafts lifelike imagery from simple textual prompts. DALL-E 2, on the other hand, is a speed painter, weaving a tale from a mosaic of random dots using a technique known as “diffusion”. The end result? A flurry of unique images generated in mere seconds.
High-Def, Lifelike ImageryWhile DALL-E 1’s creations were charmingly cartoonish against minimalist backdrops, DALL-E 2 is the digital Rembrandt of our time, transforming ideas into detailed, high-resolution masterpieces that boast of realism and adaptability.
Simplified, Intelligent Image EditingDALL-E 1’s “inpainting” feature felt like having a personal digital assistant tidying up your images. Now, with DALL-E 2, not only do you have an assistant, but an entire creative team at your disposal. With just a simple description, it can modify an image with nuanced touches like lighting and shadows or even add new items, creating a seamless blend of AI-generated and original imagery.
Endless Iterations of ArtistryDALL-E 2’s “variations” feature is like having your very own Picasso, taking your sample image and spawning countless iterations, from near replicas to abstract impressions. Add another image, and watch as it artfully merges the most significant elements of each.
Final Thoughts
The evolution from DALL-E 1 to DALL-E 2 showcases the beautiful harmony between human creativity and artificial intelligence. These AI image generators, while still experimental, open a portal to a future where imagination knows no bounds. As we continue to explore the possibilities of these tools, the future of AI in enhancing our creative potential looks promising.
According to the Copyright Office, the public has unrestricted rights to reproduce, publish, or sell any masterpiece created by DALL-E 2. However, this could be subject to change if Congress decides to amend the law, or if the courts choose to acknowledge a copyright in AI-generated work, regardless of the Copyright Office’s current stance.
OpenAI launched their newest GPT-3.5-Turbo model in March 2023, which has since been incorporated into the well-liked ChatGPT offering.
This model currently represents the pinnacle of language model technology and is 10 times more affordable than its GPT-3 predecessors.
In a shift from older models, which processed unstructured text in a token sequence, GPT-3.5-Turbo adopts a unique method, interpreting a series of messages using a system known as Chat Markup Language, or “ChatML”.
Is GPT-3.5 free?
You can access GPT-3.5 for free by signing up for ChatGPT for free.
Is GPT-3.5 available to the public?
Yes, via OpenAI’s API & via ChatGPT.
What is the difference between GPT-3.5 and GPT-4?
OpenAI launched its advanced language model, GPT-4, on March 14, 2023. It’s an upgrade from its predecessor, GPT-3, bringing more capabilities and improvements. GPT-4 is based on the Generative Pretrained Transformer (GPT) framework, which learns to emulate human communication by processing vast amounts of text.
One of the significant upgrades in GPT-4 is the ability to process images as inputs, in addition to text. This multimodal capability allows GPT-4 to handle more complex tasks, such as image captioning, summarization, or translation. This feature gives GPT-4 a significant edge over GPT-3, which can only accept text inputs.
Although there’s no official announcement about GPT-4’s parameters, it’s suggested that the number is considerably higher than the 175 billion parameters of GPT-3. This increase in parameters allows GPT-4 to generate more coherent, contextual, and suitable text, improving its performance on complex tasks.
GPT-4 also demonstrates an enhanced performance capacity over GPT-3. It can generate intelligent and creative responses to various types of inputs, including research papers, poetry, and legal briefs. Notably, GPT-4 has a significantly lower rate of “hallucinations,” or irrelevant responses, than GPT-3. This improvement in accuracy makes GPT-4 more reliable for real-world applications like natural language processing, chatbots, and automated customer service.
GPT-4 can be accessed in a limited capacity through ChatGPT Plus, where users can interact with the model and provide feedback. The availability of GPT-4 on the free version of ChatGPT is yet to be announced. Developers interested in GPT-4 can also sign up for the commercial GPT-4 API waitlist.
In conclusion, GPT-4 has showcased significant enhancements over GPT-3, making it a clear winner. Its advanced features and improved performance herald a new era in natural language processing, opening up exciting possibilities for the AI community.
What is the difference between GPT-3.5 and GPT-3?
GPT-3 is a deep learning-based language model, released in May 2020, that can generate human-like text, including code, stories, and poems. It boasts 175 billion trainable parameters, making it the largest language model to date.
Language models are statistical tools that predict the next word(s) in a sequence based on a probability distribution. They are trained on large text corpora, such as books, news articles, and web pages, learning the patterns and relationships between words in the language. Language models are utilized in various Natural Language Processing (NLP) tasks like Part of Speech (PoS) Tagging, Machine Translation, Text Classification, Speech Recognition, Information Retrieval, News Article Generation, and Question Answering.
GPT-3.5 was unveiled with the release of ChatGPT, a general-purpose chatbot. This iteration was trained on a combination of text and code before the end of 2021, learning to understand the connections between sentences, words, and parts of words by consuming vast amounts of web content. Instead of releasing GPT-3.5 fully, OpenAI used it to develop several systems optimized for various tasks, all accessible via the OpenAI API.
One system, text-davinci-003, is said to handle complex commands better than models built on GPT-3 and produce higher quality, longer-form writing. OpenAI data scientist Jan Leike stated that text-davinci-003 and GPT-3.5 have higher human preference ratings and fewer severe limitations. Data scientists at Pepper Content have also noted that text-davinci-003 excels in comprehending context behind a request and produces better content as a result, with fewer instances of “hallucination” – creating inconsistent and factually incorrect statements.
GPT-3.5 can be accessed through the OpenAI Playground. The latest model, text-davinci-003, has improved output length, generating 65% longer responses than its predecessor.
While GPT-3.5 appears to be an improvement over GPT-3, it’s unclear exactly how, as OpenAI has not released any official information about GPT-3.5. It is speculated that the improvement could be due to the training approach used for GPT-3.5, which involved human trainers evaluating and ranking the model’s responses, with this feedback then incorporated into the model.
GPT-3.5 still suffers from the limitations of modern language models. It relies solely on statistical patterns in its training data rather than truly understanding the world. It is susceptible to “making stuff up” and its knowledge of the world beyond 2021 is limited. Also, its mechanisms to prevent toxic outputs can be bypassed.
Hey, this video is about chat GBT plugins and I now have access to this. And I’m going to show you kind of what what they’re saying in this this welcome email, along with, there’s quite a few links in here. And then we’re gonna actually go into the chat GPT plugins and see what we can do a couple of important, I’m going to click on these links, I’m gonna show you these in just a second. But for unverified plugins, you can still test them, and then maximum 15, which is quite a bit, I think, you know, even if you have a pretty sizable development team, for testing plugins, and then submitting them and I have watched a few trainings on how to actually build these. And I don’t think it’s super complicated, especially if you already have an outside piece of software that can connect into it. And this is also demonstrates that if you are a developer, make sure you put the developer option when you go to request access, because I am betting that they they’re wanting to give developers this access to this first. And I won’t go into that too much into this, let’s go to this first link. So new risks. So this one, this is sort of a, you know, a technical paper, where basically they’re calling these augmented language models, which is basically what they’re saying chat GPT now is with the plugins, because it’s not just the large language model, the large language model is actually connecting to outside the ability to skills and ability to okay, it’s augmented with reasoning skills, and the ability to use tools. This is really, really cool. You know, I don’t I think a lot of people weren’t expecting this. So this is pretty cool from Cornell, let’s go into they also did a link to their system card, you know, there where they’re acknowledging that there’s safety challenges, blah, blah, blah, you know, this document contains some content that may be disturbing or offensive, because that’s what they’re trying to, to remove. And so they’re running that test there. So we’ll send you a link to that if you want to pour through this. It’s a pretty long document, so I won’t go, I won’t go through it in full. There’s also a developer, the developer community here a forum essentially, that looks like there’s, you know, there’s not a lot of people in here already. So I think this is a pretty good opportunity, especially if you already have some software, I am actually building software that I think that this, there’d be a simple plug in that would work pretty well, with chat GPT and amplifying church EBTs capabilities along with amplifying my software’s capabilities that should come out pretty soon. Hopefully, you’ll see kind of a demo of that, within about a week or so. And this is the forum for for plug in risk registrations. It looks like some submissions we open Tuesday, April 11. So you know that I’m shooting this on April 4. And so you know, you can you know, you have about a week. So maybe we’ll we’ll try to have some first plugins in there so that we can get them in there as quickly as possible. I just sent a screenshot of that to my developer, like, hey, let’s see what we can do. I hate to sidetrack her. But you know, this is such a great opportunity if we’re ready at the time. And then there’s some documentation on on how to use the plugins. This is really cool. I think part of this is building your own plugins. But also, you can actually, supposedly, you should be able to call other people’s plugins within your own software, which is really neat. So let’s go in here and see what this looks like. I’m not going to do GPT. Four, it’s been having issues lately. So let’s just kind of walk through this process. So in summary, you know, this is saying, Hey, we may send parts of your conversation to other people. And you don’t actually tell it unnecessarily what how to exactly how to use the plug in check up, figures it out. And then it’s also like, hey, trust these plugins before you install them. So let’s check out see what we got available here. Zapier is really cool. Let’s go ahead and install this one, because that opens up a lot of different options. Okay, I had to log in. And I’m going to allow it allows turn chat to actually connect up to over 5000 apps very cool. A little pop up came up. And so let’s see, you haven’t had any actions yet. You can trigger zap your actions right inside GPT. Here, a few ideas create a new draft and Gmail. churchy people will only be able to run actions that you’ve enabled, so you can always be in control. Let’s click learn more. Okay, I’ll put a link to this below as well. There. Let me let me I’ll scan through this real quick and then I’ll let you give you the highlights. Okay, long story short, you have to add actions within Zapier first, and then when you use the plugin, it can it’ll allow you to allow some of these actions to automatically happen. So let’s just Check this out real quick let’s add this hopefully I don’t reveal any information let’s look here action so like I said this is the limited amount of things in here I thought it would be actually able to do a whole lot more so there’s very limited options is kind of surprising that it’s so limited but let’s, let’s just try this because this is pretty low, low risk. Okay, so I connected to the sheets. Let’s see what when the action is run we can have a I guess use specific value. Let’s just have a I guess I’m really curious. Spreadsheet worksheet have a I guess. Okay, that’s really interesting. Action name. Google Sheets. Optional. This can help. Alright, we’ll just do Google Sheets Add Row. Alright, let’s enable action. Let’s see what happens so add new action. Okay, so we have the action. Now we need to go into our touch up T account and we’re going to select Zapier Plugins Add me think it’s something here I’m gonna pause real quick. Let’s just try something real simple. Please create let’s do create a rap like Yoda about artificial intelligence and store the rap in Google Sheets. Let’s see how this works. I’m gonna I’m gonna pause well let’s see what Okay, so now let’s store this rap and so did the rap. Hip. I’ll let you read that pause if you want and it’s only using Jeep This is GPT 3.5. So not GPT for a good GPT for has been having some issues lately. To do that, you’ll need some information about the spreadsheet. I’ll need information where you’d like to Could you please provide the name of the spreadsheet the name of the worksheet where you’d like to add the rep. Okay, so I created this new spreadsheet Yoda isms and put rap in here. Okay, so Yoda isms and raps. Let’s see. So, spreadsheet equals Yoda isms. Worksheet equals raps reps. See what it does. Using Zapier. Not going to pause. So I want to I want you to see potentially how long it takes because the longer this takes, the less practical it is. unrecognized error seems that there was an issue with adding the wraps of Google Sheets document, I apologize for the inconvenience. Let’s try a different approach. It will format the rap as a single string and tend to add in a single in a specified spreadsheet which again, here we go. Let’s try adding again. Oh, wow. It’s so it failed. And then it’s going to try again. So I’ll just kind of rap a little bit about the fact that the actions within Zapier are extremely limited right now. It’s kind of disappointing, to be honest. And also now it seems like there is a there’s there, it’s having trouble. It seems there’s an issue, adding the wrap to it. Let’s see if it actually added it. So it didn’t add it. Okay. Well, I’m not going to try this again. Let’s Let’s do well, is it still running it? unrecognized core kwargs arrow wrap. I don’t know why it has trouble with wrap. Worksheet equals wraps. I did create an action that adds a row. I wonder if there’s a way saved. Let’s click here. Let’s see if it didn’t save it. Okay, so Google Sheets Add Row. Maybe I need to be more specific. So have a I guess. Maybe we try maybe just try again. You know, let’s let’s just let it beat let’s just say a little there’s a little bit of a clunky process. So maybe some of the things saying oh my god, this is amazing. And it’s immediately going to be able to do this X, Y and Z and maybe not so much. So let’s let’s find a few more that are interesting. I like the the Wolfram Alpha One. Not really interested. Okay to turn any 20 minutes from Metamagic Okay, I have kids so that’s interesting. I’m not I’m not really interested in these. Well, maybe this would be interesting. Any of these other ones I don’t think it’s really very practical to right now to maybe book via Expedia and kayak especially OpenTable I’m very very suspicious of that. Let’s see what else we have. Learn how to say anything another language was speak you’re a power a powered that’s interesting. See, I don’t you know, the fact that there’s like no Amazon in here, or you know, major companies. I’m not sure it will go down and stuff. Listen, see what happens because I’m not really interested in these other ones, although, you know, if you’re and then here we go develop your own install verify, and unverified plugin developer and plugins about plugins. So let’s try like. So you can enable multiple ones, because you can only have three enabled at once. So let’s just see if we, if we do what’s, please tell me what a 37 cubed is. Let’s actually do 37 times 37 times 37. So it’s 50,000. Okay, so it used Wolfram good. That’s good, because that’s one of the biggest issues. So I didn’t have to tell it to use Wolfram, I just had to enable the plugin. Let’s see. So meal, curing the wisdom village, I don’t understand how this works. Let’s turn off Zapier because it sucked. And we’ll do that. So to give parents what’s actually Meelo family AI, let me look that up real quick. So found this site, it’s very light on information, maybe I’ll go ahead and join the waitlist to show that it can work. I’ll do that later. Because it looks like I’ll have to take a couple minutes. And I know that if you don’t if you’re not interested in this, if you don’t have kids, it’s you know, this seems actually much less developed than I expected. I’m not really sure how to enable this plugin. But like, let’s see. Tell me about Meelo family and let’s see if I just kind of ask you, please, we always want to be polite to our new robot overlords. Okay, so it doesn’t know about I mean, I guess that makes sense. Maybe let me try a different search on Google. So I searched Meelo family AI chat GPT commands? Well, yeah, I’m not really there’s not really a lot of information on this. So let’s just maybe say, let’s try a new conversation. Let’s just try. Please suggest fun games with my kids. Let’s see if it uses Meelo here. Just by mentioning, mentioning kids didn’t use it. So we’re gonna we’re gonna go to a different one. Let’s try the shopping thing. What some top rated office chairs. Please tell me about top rated again, this is these are this is not great prompt engineering. The whole point here is just to see okay, there we go. So it’s actually integrating top, okay. Expected numbers to be string error, use sharp. Okay, if I click this link, what happens? So it’s going to a m t, let me pull this down for you. A N T. I’ve never heard of this store. So little sass there. It’s a New Zealand dollars. Okay. So what I’m seeing right now pretty consistently is that this is now maybe this is, you know, they’re releasing developers. So and they’re probably actively working on this right now. So I would expect that when they do release this outside of developers that it should be a lot better. And so you know, there’s kind of this misconception of this, this bad thought process that I see a lot where like, it doesn’t work now. Therefore, what work? The one exception to this was his Wolfram Alpha, let’s take a look at some of these other places. So I’ll just leave the URL open. So you can see I trying to think if I’ve heard of any of these stores, I have not. So I don’t think this shop plugin is very useful. Let’s go in here. Let’s go new chat. I’m gonna back to the plugins fretless if there’s anything else that I might want to check out, Okay, one more maybe. And if there’s anything else on here that you see that you want me to try out, let me know. Let’s try this one. And we’re going to turn off this because, well, let’s actually turn off sharp and turn on speak. Alright. I’m here, okay. Please tell me how to say okay, please tell me how to say check several chats up these plugins need some work in Spanish. Wonderful. Oh, there we go. Okay, so I’m not sure if If it’s actually going to speak to me, we’ll see. See how long this takes. Doo doo doo doo. I’ll give you the music while go doo doo doo doo doo doo doo. Okay, I thought it would actually speak it for me. Ha huh? So click that what does it do? Doesn’t? That’s really interesting. Let’s go back to the store, see what it says? Learn how to say anything in other language your app. I mean, couldn’t GPT four already do this? I don’t really understand what the purpose is if if this was like a speaking thing, that sounded pretty cool. And where is the plugin, there’s supposed to be a plugin that allows you to the plugins built by by opening if there’s the data plug in that that was actually the I think they call it the program and remember which one they called it the programming plug in, but it allowed it to analyze large portions of data. I don’t see that in here. That one was the one that I thought would be the most useful. Other than that, I don’t really that’s kind of interesting. So maybe that’s just maybe they’re, they’re tuning it and they took it off the store for right now. But, you know, let me know if there’s a plug in that you want me to try maybe we’ll shoot a few more videos or if you have any comments or questions or thoughts or tests that you want me to run with the current plugins happy to do so. If you liked this, please like and subscribe for more videos like this. If you don’t like this, then you can go away in politely please. I’ll see you in another life. Thanks for watching and have a great day. Bye
Yes, GPT-4 is available for the public to use via ChatGPT Plus ($20/m).
Can I use GPT-4 for free?
No, currently, there is no way to access GPT-4 directly for free. There are, however, some indirect ways to access GPT-4 for free. The places where this is available changes frequently, so we will not link directly. We suggest you Google something like, “GPT-4 Free access.”
What can GPT-4 do?
Enhanced language understanding:
GPT-4 could demonstrate even better context understanding, allowing it to generate more coherent, contextually relevant, and grammatically accurate text than GPT-3.
Improved few-shot learning:
Building on GPT-3’s few-shot learning capabilities, GPT-4 might require even fewer examples to understand and adapt to new tasks, leading to more efficient fine-tuning and adaptation.
Task versatility:
GPT-4 could be capable of tackling a broader range of tasks with higher accuracy, such as summarization, translation, question-answering, code generation, sentiment analysis, and more.
Longer context retention:
GPT-4 might be able to handle longer text passages more effectively, leading to better performance in tasks that require understanding and manipulating extensive context.
Enhanced commonsense reasoning:
GPT-4 could exhibit improved commonsense reasoning and inference capabilities, which would enable it to generate more plausible and sensible responses in a variety of situations.
Reduced biases:
GPT-4 might benefit from continued research in AI ethics, leading to reduced biases in its responses and a better understanding of contextually appropriate language.
It’s important to note that this is speculative and based on the progress observed in the GPT series and AI research in general. The actual capabilities of GPT-4, if and when it is developed, might differ from these predictions.
It is unlikely that GPT-4, or any AI-driven language model in the near future, would completely replace programmers.
While advanced language models could assist programmers by automating repetitive tasks, providing code suggestions, and aiding in debugging, there are several reasons why they are unlikely to replace human programmers:
Human expertise and creativity:
AI models may provide support and suggestions, but they cannot replace the creativity, critical thinking, and problem-solving skills that human programmers possess. Innovative solutions often require human intuition and expertise that AI models have not yet been able to replicate.
Complex projects:
AI-generated code snippets can be helpful for simple tasks, but complex projects require a deep understanding of the problem domain, system architecture, and trade-offs involved in design decisions. Human programmers have the ability to plan, strategize, and adapt to the unique challenges posed by intricate projects, which AI models are not yet capable of handling.
Collaboration and communication:
Effective software development often involves working closely with other professionals, such as product managers, designers, and other stakeholders. Programmers need to communicate effectively and collaborate on various aspects of a project, which is a skill set that AI models currently lack.
Limitations in AI understanding:
Although advanced language models like GPT-3 have made significant progress in language understanding, they still have limitations in terms of context understanding and may not always provide accurate or appropriate suggestions. Human programmers are better equipped to discern when a solution is viable or when further refinement is needed.
In conclusion, while GPT-4 model might serve as a valuable tool to assist programmers, it is unlikely to replace human programmers entirely. Instead, AI models will likely complement the work of programmers, allowing them to focus on more creative and strategic aspects of software development.
