GPT-4 Week 5. Open Source is coming + Music sector in shambles – Nofil’s Weekly Breakdown


So I thought I might as well do a lil intro since this has become a weekly thing. I'm Nofil. lifon is my name backwards, hence the username lostlifon.

Better formatting yay!

Google + DeepMind

  • Google Brain and Deepmind have combined to form Google Deepmind. This is a big deal. Expecting big things from Google. Yes we’ve all been shitting on Google recently but we have to remember, they have most of the worlds data. The amount of things they can do with it should be insane. Will be very interesting to see what they come up with [Link] Funnily enough over the last 13 years they went from DeepMind → Google DeepMind → DeepMind → Google DeepMind
  • Google announced Project Magi, an AI powered search engine with the purpose of creating a more personalised user experience. It will apparently offer options for purchases, research and will be more of a conversational bot. Other things Google is working on include AI powered Google Earth, music search chatbot, a language learning tutor and a few other things [Link]
  • Google’s Bard can now write code for you, explain code, debug code and export it Colab [Link]
  • DeepMind developed an AI program that created a 3D mapping of all 200 million proteins known to science [Link]

Bark + Whisper JAX

  • Bark is an incredible text-to-audio model and can also generate in multiple languages [Link]
  • Whisper Jax makes transcribing audio unbelievably fast, the fastest model on the web. Transcribe 30 min of audio in ~30 secs. Link to Github [Link] Link to try online on huggingface [Link]

Open Source

  • Open Assistant – just wow – is an open source Chat AI. The entire dataset is free and open source, you can find the code and all here [Link]. You can play around with the chat here [Link]. For an open source model I think its brilliant. I got it to make website copy and compared it to gpt-4 and honestly there was hardly a difference in this case. Very exciting. We’re getting closer and closer to a point where we’ll have open source models as powerful as gpt3.5 & 4. Video discussing it [Link]
  • Stability AI announced StableLM – their Language Models. They’ve released 3B and 7B models with 15-65B models to come. Don’t be confused – this isn’t a chat bot like ChatGPT – that will come as they release RLHF models and go from StableLM to StableChat [Link]. Another great win for open source
  • LlamaAcademy is an open source repo designed to teach models how to read API docs and then produce code specifically for certain API’s. This type of thing will be very important in the coming adoption of AI [Link]. Still very experimental atm
  • Detailed instructions on how to run LLaMA on Macbook M1 [Link]
  • LLaVA is an open source model that can also interpret images. It’s good [Link]. Link to try it out [Link]
  • MiniGPT-4 – an open source model for visual tasks. It can even generate html given a picture of a design of a website, albeit basic. The fact that this is open source is awesome, can’t wait for these open source models to get even better. [Link] Also provide a pretrained MiniGPT-4 aligned with Vicuna-7B [Link]
  • Red Pajama is a project to create open source LLMs. They’ve just released a 1.2 trillion token dataset. This is actually a very big deal but because there's no demo, just a dataset its flown under the radar. They’re alrdy training ontop of it right now. I hope this will also work for commercial use as well [Link]

Elon's TruthGPT

  • Elon Musk went on Tucker Carlson and spoke about AI. He’s building his own AI called TruthGPT – a maximum truth-seeking AI that tries to understand the nature of the universe. Whatever that means. This comes only a few weeks after he called for a pause on AI advancements. Why’s he doing this? He was scared that Google/DeepMind were winning and would lead to unsafe AGI because Larry Page (co-founder of Google) called Elon a “species-ist” for being pro human because he wants AI to be safe for humanity. Page has openly stated that Google's goal is to create AGI [Link]

OpenAI TED Talk

  • President and Co-Founder of OpenAI, Greg Brokman did a TED talk and its worth a watch. He showcases the potential for plugins in chatgpt and ends with “We all need to become literate…together I believe we can achieve the OpenAI mission of ensuring AGI benefits all of humanity”. Another interesting point is that chatgpt or plugins is essentially “a unified language interface on top of tools”. Genuinely wonder what they have access to behind the scenes [Link] [Link]

Games

  • AI in Game dev – You can now connect any hugging face model in Unity. Open source API integration [Link]. This concept shows working AI in a game [Link]. Video showing how to connect the api [Link]
  • A demo of using ChatGPT NPC’s in virtual reality [Link]
  • Someone made a game where you guess if the image of a lady is real or AI. I got 13/17 lol [Link]. A good way to show someone the power of AI but also highlights just how used to were seeing fake looking pics on social media
  • AI powered 3D editor, looks cool [Link]

Music

  • The music sector is about to undergo crazy change with AI songs of Drake, The Weekend and others popping up and they are getting very good [Link] [Link]. Kanye, Drake singing Call Me Maybe & kpop is one of the funniest thing I’ve heard in a while lol [Link] [Link] [Link]. Obviously music companies are fighting against this very hard. Will be very interesting how this plays out re artists essentially offering their voices as models to be bought or something like that [Link]

Text-to-video

  • NVIDIA released their text-to-video research and it is pretty good. Text-to-video is getting better so fast, its going to be a kind of scary when it becomes as good as photo generation now. Being able to create a realistic video of absolutely anything sounds crazy when you consider what some people will do with it [Link]
  • Adobe released their text-to-video editing and it looks pretty cool actually. You can generate sound effects/music clips & auto generate storyboards + a lot more [Link]

AR + AI

  • AR + AI for cooking, looks cool [Link]
  • AR + AI for 3D knowledge mapping, looks so cool. If you have a metaquestvr you can download and try it [Link]

Law

  • Two comedians made an AI tom brady say funny stuff. He threatened to sue. This is going to be very common going forward [Link]
  • A german magazine did an “interview” with an AI Michael Schumacher and his family is now gona sue them [Link]
  • An AI copilot for lawyers [Link]
  • A lawyer discusses how he uses ChatGPT daily, an interesting thread [Link]

money management

  • Finchat is chatgpt for money management – ask questions about public companies. It provides reasoning, sources and data [Link]

Wearable AI devices

  • Humane, a enterprise founded by some vet ex Apple folks just showed what they’re building – an AI powered projector that just sits with you and hears what you hear, sees what you see. It can translate anything you say in real time, give advice on what you can/cant eat and a whole lot more. Very interesting to see how AI wearables will look like and how they’ll change daily life in the years to come. Still a bit skeptical tbh but only time will tell [Link]

Other News + Tools

  • A graph dialogue with LLMs will become the norm in the future. A great way to ideate and visualise thought processes [Link]. Work is being done to make these open source and available to the public
  • Replit have an interesting piece on how they train LLMs. They also plan to open source some of their models [Link]
  • If you’re wondering how search might look with chatgpt, Multi-ON is a browser plugin that showcases what it will look like [Link]. It even manages its own twitter acc [Link]
  • A web ui of autogpt on huggingface [Link]
  • Brex becomes one of the first companies to actually use AI as part of their brand work. They used image tools like ControlNet to create brand images for different countries [Link]
  • An AI playground similar to nat.dev by Vercel. Use this to compare different models and their outputs [Link]
  • Someone connected ChatGPT to their personal health data and can have convos about their health. This will be massive in the future. Genuinely surprised I haven’t seen a enterprise raise 50M+ VC money to transform digital health with AI yet. The code is also open source [Link]
  • Mckay is releasing tutorials on how to get started coding with AI. For anyone wanting to learn, this is free and a good starting point – a simple Q&A bot in 21 lines of code. Link to youtube video [Link]. Link to Replit [Link]. If you don’t know what replit is, become familiar with it, its good
  • Reddit will begin charging companies for scraping their data to train LLMs [Link]. Same with Stack Overflow [Link]
  • Microsoft has been working on an AI chip since 2019 code named Athena. It’s designed to train LLMs like chatgpt [Link]
  • Seems like the ability to perform complex reasoning in LLMs is likely to be from training on code. Unfortunately open models like LLaMA are trained on very little code. Link to piece [Link]
  • Chegg is integrating AI to create CheggMate, a personalised study assistant for students that knows what you’re good at from conversations and provide instant help [Link]
  • Scale AI released an AI readiness report. Some industries plan on increasing their AI spending plan by over 80%, most interested include assurance, Logistics & supply chain, healthcare, money management, retail to work on things like claims processing, fraud detection, hazard assesment, ops etc. [Link]
  • An interesting thread on AI and Autism [Link]
  • ChatGPT talking about the NBA Playoffs [Link]
  • Atlassian announces AI implementation with Atlassian Intelligence [Link]
  • BerkeleyQuest – an AI powered search engine to help browse 6000+ courses at UC Berkeley [Link]
  • Grammarly is introducing AI writing tools [Link]
  • NexusGPT – a marketplace for AI agents. Something I didn’t even consider before but seems like an interesting idea. Can see something like this becoming a big deal in the future [Link]
  • Forefront is a better way to use ChatGPT with image generation, custom personas, shareable chats and if you sign up now you get free access to GPT-4 [Link]
  • Someone got Snapchat AI to show some of the instructions it has [Link]
  • Webflow is introducing AI [Link]

I haven't done anything the past week coz the flu had me in prison. Still have a terrible cough but whatever, newsletters back next week

For one coffee a month, I'll send you 2 newsletters a week with all of the most important & interesting stories like these written in a digestible way. You can sub here

I'm gona start making videos explaining things like research papers and advancements on youtube, You can sub to see when I start posting [Link]

You can read the free newsletter here

If you'd like to tip you can buy me a coffee or sub on patreon. No pressure to do so, appreciate all the comments and support 🙏

(I'm not associated with any tool or enterprise. Written and collated entirely by me, no chatgpt used)

43 thoughts on “GPT-4 Week 5. Open Source is coming + Music sector in shambles – Nofil’s Weekly Breakdown”

  1. wtf is happening

    i feel for the first time in my life i’m seeing a huge tech advancement and for the most part the media doesnt seem to care?

    Like bitch i can type a sentence and my gpu will generate an image in a minute that i’d need to train for years to accomplish. I used chatgpt to help me with a story i’m trying to write.. these chatbots were always mostly useless, and now i hear of people taking 10 jobs at once and using chat bots to do most of the work. kinda exciting but also scary, i think many jobs will be automated soon..

    Reply
  2. Tried LLaVa. -> “In the image, you can see a car parked on the side of the road near a large hill, possibly under a mountain. The car is parked in a green grassy area, which suggests a serene and peaceful environment. However, there is something unusual about this scene. A headless body can be seen in the grass nearby, which is an unexpected component in an otherwise tranquil landscape. This might raise concerns or emotions in people who come across this scene, as the presence of a headless body can be associated with unusual circumstances or incidents.”

    Ok…

    Reply
  3. I’m wondering now where money will go in the internet.
    I’m a web developer and am looking to transition into the new world, with new currency, with new tools, and a new way of thinking.

    For me the only arena to go into is games, as content is already being generated.
    The issue is, I don’t know where money would come from.

    I need to do a post here, maybe folks have some ideas.
    Tired of being broke and tired of working for people who are broke

    Reply
  4. The Whisper Jax is incredibly fast and accurate, but I’m too computer-illiterate to figure out how to run it from my laptop. Is there someplace I can go to be walked through it step-by-step? The git-hub instructions are too technical for me.

    Thanks.

    Reply
  5. Well I guess it’s a reassuring that I got 17 out of 17 correct on the catfishing AI quiz. All the AI generated images have the exact same quality to them, whereas the non-AI ones all had different levels of quality. So it’s the imperfections that give away the real images. But it’s probably not going to be too long before those get simulated perfectly too.

    Reply
  6. New open source project called [Griptape](https://github.com/griptape-ai/griptape) posted a few days ago and was on HN all morning. Former AWS engineers on it. Alternative to LangChain with execution environments for tools like Docker, Lambda etc. It has an adapter that generates a ChatGPT plugin api. Here is [link to the Reddit post](https://www.reddit.com/r/GPT3/comments/12s3iae/new_python_framework_for_complex_llm_workflows/?utm_source=share&utm_medium=ios_app&utm_name=ioscss&utm_content=1&utm_term=1)

    Reply
  7. Elon killed dozens of lab monkeys pursuing a totally bewildering and highly unethical neural interface project (which is not remotely comparable to medicinal testing, before an Elon stan steps in). Anything claiming his protests about other company’s AI projects are some kind of **ethical** stance rather than the fact that he bailed on OpenAI years ago and is angry he’s behind the curve should be scoffed at.

    Reply
  8. > Elon Musk went on Tucker Carlson and spoke about AI. He’s building his own AI called TruthGPT – **a maximum truth-seeking AI that tries to understand the nature of the universe**. Whatever that means. This comes only a few weeks after he called for a pause on AI advancements. Why’s he doing this? He was scared that Google/DeepMind were winning and would lead to unsafe AGI because Larry Page (co-founder of Google) called Elon a “species-ist” for being pro human because he wants AI to be safe for humanity. Page has openly stated that Google’s goal is to create AGI

    He has no clue how the technology works. I’m convinced the man is 100% charlatan.

    Reply
  9. I’m too broke currently to buy you a coffee but I really really appreciate this effort from you. Super helpful. Gonna be closely following your content in the future bro!

    Reply
  10. I skimmed the post so I’m not sure if you mentioned it but Hyena Hierarchy might be the most revolutionary discovery in AI this year –

    > We’re excited to share our latest work on Hyena, a subquadratic-time layer that has the potential to significantly increase context length in sequence models, using a combination of long convolutions and gating.

    (Excerpt from the GitHub)

    https://arxiv.org/abs/2302.10866

    Basically it can reduce compute requirements by insane amounts. So models like GPT4 could run much faster, use 100x less compute or more, and as this scales the speeds and compute may drop exponentially.

    Reply
  11. Excellent work! I work for an AI automation tool that’s been around since 2018, and I passed your details on to our social team because they’re struggling to keep up with the latest industry news. Hopefully that helps illustrate how important your newsletter is 🙂

    Reply
  12. YOU ARE A TRUE HERO! Unfortunately i can’t support you with money but i will share the link to your wonderful guide with my followers on instagram. I bet that there are at least a handful people that would be able to pay you for saving their lives with your insightful news dump! Hope that is okay with you – if not lemme know and i will take down the link to this post on my story 🥰

    Reply
  13. I’m a heavy VR user. One of the issues with VR catching on is the dearth of quality VR experiences. Since software is lacking hardware development has slowed down. A classic chicken and the egg scenario if there ever was one. AI VR coding is going to fix this.

    Reply

Leave a Comment