💰 OpenAI to soar to $1T valuation, researchers make AI mimic butterflies, and world's first AI-powered driverless drifting car

Is AI getting too ‘good’? We’re rapidly moving into a reality where ‘mixture of experts’ models, deep-fake video and audio technology, and open-source tools for virtually any use case are all readily available, and the implications are staggering. Moving into election season in the US, the recent demos from HeyGen - where their photorealistic avatars can move and speak are fooling people everywhere into believing they’re real, and OpenAI just announced their new Voice Engine, which all truly begs the question of “how will we know what’s real and not”?

I love sharing new tools, tutorials, prompts, and growth hacks here in AI Logs, where today we celebrate our 25th issue. But we also need to use my platform to help facilitate conversations around how we keep AI “good” as it keeps getting “better.” My assessment has been that we need to tie any and everything AI-generated to an immutable blockchain using smart contracts to have a digital chain of custody and guardrails in place. After years of saying this, I’m finally meeting others who believe the same and think it is as critical as I do. Some startups are also building AI on-chain, as we’re doing at my stealth startup.

We only have one chance to get this right!

But in other AI news for the week, a headline that you may have seen floating around is that “Claude beats GPT” in certain benchmarks, which I don’t want to overstate, as in my experience ChatGPT and Google Gemini are still my go-to’s as they do much more of what I need my AI to than Claude, and are more capable for my use cases (I also love playing with Mixtral, Llama and Gemma in Groq). However, I do think the headline is noteworthy because, in my assessment, it will almost certainly motivate OpenAI to release some new capabilities to "regain top spot" (perhaps 4.5? Fingers crossed…) and it will prove what most of us know, which is that the tech is already significantly ahead of what's widely available.

The other big news from the AI world in this issue is that OpenAI and Microsoft have announced plans to build a $100B supercomputer called “Stargate,” likely harnessing the power of thousands of both GPU and LPU chips in the quest for adequate compute power. Although the proposed timeline goes out to 2028, there will likely be some huge developments in this quest much sooner that could have profound implications. I still think LPUs are the disruptive game changer in compute that we all have known we need. “Watch this space…”.

Did a friend forward this e-mail to you?

IE₊ SUPPORT INTERESTING ENGINEERING
_{Invest In Science And Engineering}

Enjoy exclusive access to the forefront of AI content, highlighting trends and news that shape the future. Join a community passionate about AI, delve into the latest AI breakthroughs, and be informed with our AI-focused weekly premium newsletters. With IE+, AI reporting goes beyond the ordinary - and it is Ad-Free.

NEWS

• 🦋 Researchers make AI mimic butterflies to ensure multi-sensory decision-making

  To make AI as smart as butterflies, a team of Penn State researchers created a multi-sensory AI platform.
  

• 🏎️ China reveals world’s first AI-powered driverless drifting car 

  Geely showcased a test vehicle that performed fully unmanned autonomous driving and drifting on snow and ice.
  

• 💻 Microsoft and OpenAI pledge $100 billion for ‘Stargate’ supercomputer facility

  This project will take six years to complete, and also lead to the development of an advanced AI supercomputer dubbed Stargate, which is scheduled to be operational by 2028.

MUST READ

OpenAI to soar to $1T valuation, ex-Google China president says

Making a bold claim at the Fortune Innovation Forum in Hong Kong, Kai-Fu Lee, chairman of Sinovation Ventures and former Google China president, has predicted that OpenAI could soar to a valuation of one trillion dollars soon.

While the company, founded just eight years ago and led by entrepreneur Sam Altman, remains a loss-making venture, savvy investors valued it at over $80 billion in its latest fundraising round.

Lee highlighted OpenAI’s incredible rise, positioning it as a game-changer in the tech industry.

PROMPT OF THE WEEK

I’ve discussed things like “no yapping” and “proceed” in previous prompts of the week that many have told me are super beneficial to them as well (I highly recommend going back through previous issues of AI Logs as they’re still relevant, and compiled can be quite a level-up for those wanting to really learn AI). This week’s prompt of the week is in a similar vein; namely, when asking your AI to do a bigger project for you, it likely tells you that it will take several hours or even days, but it doesn’t actually do that or have a concept of time. So the hack is this:

Prompt it that it’s been that amount of time, and/or use its lack of comprehension of time to your advantage. When it says something like “I will be working diligently on that and will have the task completed within 2 days”, I immediately prompt back telling it that’s it’s been 2 days, and ask for the deliverables, at which point it should tell you that you’re right, and provide the output you’re expecting and had been told to wait on otherwise. You can just keep telling it that it’s been the time it said, or that it’s a day late, etc., and it will keep moving forward with the project in real time.

I do expect that as we see more robust autonomous agents capable of more robust and onerous tasks in the coming months that there will be many times that the AI does, in fact, require some time (unless it’s using Groq and/or their LPU chips!), but for now this prompt hack has helped me a lot.

AI PICTURE OF THE WEEK

TUTORIAL

This week I posted on LinkedIn about UBI (universal basic income), and wanted a graphic of Bart Simpson writing “We Need UBI” hundreds of times on a blackboard. The first one was fairly bad, but I used the following for prompts and logic, and the final result (only 2 prompts, I could have done further and got it even more dialed in) turned out quite good.

The words were perfect on the first attempt and not all perfect in the second, but it got me so much closer to my ideal output, and recreated a pop culture icon that is likely supposed to be protected by copyright etc., so I was pretty pleased. I encourage you to glean what you can from this to apply to your use cases as well (the tutorial is more in how I got it to go from bad to good almost immediately and effortlessly), as many will say AI (here I use DALL-E in ChatGPT) can’t deliver outputs like this (see image of week) but to that, I offer the following:

First prompt: “Create an image of Bart Simpson writing “We Need UBI” on a blackboard hundreds of times”.

ChatGPT: [Bad Image Generated] “How does this image align with your vision?”

Second prompt: “Well it doesn’t look a lot like Bart Simpson and it says it 4 times not hundreds, but it’s a good start. Please research the iconic memes of Bart Simpson writing the same thing over and over and over on a chalk board and try to recreate that with him writing “We Need UBI”.

ChatGPT: “The iconic Bart Simpson chalkboard scene, a hallmark of "The Simpsons" opening sequences, has been widely parodied and meme-ified since its inception. This scene typically features Bart writing lines on a chalkboard as a form of punishment, with the content often reflecting humorous or satirical messages. The first known customization tool for creating these Bart Simpson chalkboard images appeared online in 2007, allowing users to generate images with their own text on the chalkboard. This sparked a trend of custom chalkboard images that spread across various platforms, including Newgrounds, Ebaumsworld, and Tumblr, where users shared their own humorous or pointed messages in the style of Bart's chalkboard gags oai_citation:1,Bart Simpson's Chalkboard Parodies | Know Your Meme. The chalkboard gags themselves have been a mix of humor, social commentary, and even poignant acknowledgments of current events. For example, during the early stages of the COVID-19 pandemic in 2020, an episode featured a chalkboard gag that simply read "School online," reflecting the global shift to remote learning oai_citation:2,Bart's Most Savage Chalkboard Jokes From The Simpsons. To recreate an image in the spirit of this meme with Bart writing "We need UBI" hundreds of times would involve capturing the essence of these gags: the repetitive nature of the punishment, the character's begrudging compliance, and the chalkboard filled to the brim with the repeated phrase. This context provides a rich backdrop for an image that would resonate with fans of the show and the meme culture surrounding the chalkboard gags.

Third prompt: “Proceed”

ChatGPT: [New Image, much better (see Image of the Week)] “ImageHow does this revised version meet your expectations?”

Fourth prompt: “Much better”

ChatGPT: “I'm glad to hear that! If you have any more requests or need further assistance, feel free to let me know.”

TOOLS OF THE WEEK

🗣️ Hume — this is an evolution of sorts for AI, and has been making a lot of headlines this week, but if you haven’t already seen and played with it, Hume is an “empathetic” conversational AI that can understand your emotions and inflections, mirror or respond accordingly audibly with on-sentiment emotions and cadence. It is intended to be like a virtual counselor, friend, therapist etc. Even as OpenAI teases their new voice model, and Inflection AI deprecates their AI, “Pi”, an earlier “emotionally intelligent” AI (stood for “personal intelligence”), this Hume tech seems to hold a lot of promise and potential. It’s free to use — I highly recommend checking it out.

🤖 OpenDevin (GitHub) — last week I featured Devika as an alternative to Devin: both are agentic AI capable of programming, testing, and potentially even ‘active reasoning’. Devika and Devin both have a lot of potential, although they’re both early products, but this new one, OpenDevin, is available on GitHub and seems to be the most functional and powerful yet. It has an amazing bifurcated interface, is super responsive, and codes and tests really well. Definitely worth checking out if you’re either technical or want to learn to be a no-code techie.

🧐 DBRX by Databricks (HuggingFace) is an open, transformer based mixture of experts model (MoE) trained using next-token prediction, pretrained on 12T (trillion!) tokens of “carefully curated data” and is pretty remarkable as it allows users to create our own LLMs as well as AI apps. This is likely where the industry standards are heading, and well worth familiarizing yourself with.

🎲 ChatDev is a cool new tool to “build your AI town” with a game mode, chat mode, and “promptIDE”, allowing users to build their own agents. I like that it’s gamified, simple, and although it’s very early/new, it’s hard not to like the direction they’re headed in.