💰 Jony Ive and Sam Altman seek $1 billion for the 'iPhone of AI', and Robot dogs train for moon rescue missions

Hello all, I'm thrilled to announce my live workshop on Interesting Engineering Academy. The 'AI Design Studio' is a fully interactive online course customized for non-coders and seasoned programmers eager to step up their interactions with AI in software development and design.

As a special treat for our AI Logs subscribers, we offer an exclusive discount. Use the code AILOGS50 to snag an extra 50 percent off. So, what are you waiting for? Get the first course now.

This past week has me thinking that we’re potentially both closer to ChatGPT 4.5 and potentially even AGI than most ‘experts’ and enthusiasts alike seem to believe — and while I’ve said both here in previous issues of AI Logs, I have more conviction than ever now. Here’s why:

First things first, a new feature was released in ChatGPT, namely, the ability to edit images generated in DALL•E, which is a huge deal. Beyond just being an amazing functionality, typically, when new features are being rolled out, it’s indicative of a new model not being far behind it. Secondly, with other AIs such as Claude reportedly outperforming ChatGPT in certain benchmarks, it gives the industry leader, OpenAI, motivation to show what it’s really capable of — competition is a worthy motivator.

But finally, in regards to the potential of GPT 4.5, I had a breakthrough in prompting this week that had me beyond stoked, although I haven’t been able to replicate it since (which is further proof positive that an update may be imminent). I prompted ChatGPT to prompt itself, and to deliver results while still working on other tasks, and to prompt itself and to keep working autonomously, without me, and for a full project lasting nearly an hour — it DID. I was blown away. Autonomous Agents have come a long way, and I’ve featured MultiOn and others here in previous issues, but to see it inside of ChatGPT, and the results it was able to achieve, were unreal.

Although I’ve been unable to replicate it (it now tells me that isn’t something it’s ethically able to do), I have the thread, the final results and witnessed it with my own eyes, and this, coupled with the above data points, rumors I’ve seen circulating etc., have me feeling fairly confident that the only way we wouldn’t see GPT 4.5 soon is if the board determines it’s too powerful for the public, or that they’re so close to releasing GPT 5 yet, which I doubt, to justify releasing a 4.5 model.

So I’ve addressed the potential of ChatGPT 4.5 being near, but what about the claim of AGI being closer than some suspect? Well, regular readers know that I’m bullish on many models in addition to LLMs (language) such as LWM (world), LAM (action), LNM (nature), and more, and highlighted Meta’s Yann LaCun saying that “large language models alone” can’t achieve AGI. But I speculate that these models together, particularly with autonomous agents and other models recently launched specific to robots, may get us really close; like: really, really close. Now we just need more compute, which Groq and their new LPU chips may well have solved for.

There’s so much else to unpack this week that I could write another ten paragraphs here, but in the name of brevity, let’s jump into this week’s prompts, tools, tutorial, and news; I’ll try to incorporate as much of the week’s headlines there too as possible.

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

• 🐕 Robot dogs drill to train for moon rescue missions, navigating craters

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

• 🗳️ China could deploy AI to disrupt elections in US, India: Microsoft warns

  Presidential elections in Taiwan were a dry run and China could sharpen its attack in high profile elections lined up in 2024.
  

• 💻 Why did whisper take a million hours of YouTube videos?

  As per reports, the OpenAI team illegally used more than one million hours of YouTube videos, here is why.

MUST READ

Jony Ive and Sam Altman seek $1 billion for the “iPhone of AI”

In September 2023, reports emerged of former Apple design guru Jony Ive joining forces with OpenAI CEO Sam Altman to develop the “iPhone of AI.”

The duo held brainstorming sessions at the designer’s San Francisco studio about a new consumer product centered on OpenAI’s technology.

SoftBank CEO Masayoshi Son was reportedly involved in some of the discussions, “pitching a central role for ARM, as well as offering financial backing.” The Japanese multinational investment company holds a 90 percent stake in the chip manufacturer.

PROMPT OF THE WEEK

I’m tempted to share the prompt that got ChatGPT to run autonomously, but since it’s no longer working and this section is meant to be actionable, I’ll switch gears (many of the elements of the “autonomous” prompt are previous “prompts of the week” here anyhow).

So instead I’ll offer an ‘anti-prompt’, as it’s how to voice-speak with your AI, similarly to Hume or, until recently, Pi, rather than text-prompt it. But it is remarkable and noteworthy how GREAT the results are getting when we speak to our AI (ChatGPT, Gemini, etc.), and renders the skill of “prompting” virtually obsolete, although the same underlying principles remain.

You can talk to most AIs using the microphone icon, and can have it read its responses back audibly after it types it, but several let you actually have a voice conversation, and ChatGPT is a great one for this. Click the headphone icon to the right of the chatbox, and you can start asking it questions, practicing sales pitches or interviews, having it effectively act as your teacher, or so much more.

A few great use cases for this are practicing a keynote and asking for feedback, learning a new language and speaking in that foreign language at a level that you pushes you to the next one, polishing a sales pitch, learning history and much more.

Why this works — we often speak faster than we type and don’t overthink it as much, we also often learn better by hearing than by reading. It also makes your AI more like a companion or mentor. If you haven’t tried this previously, I highly recommend it.

AI PICTURE OF THE WEEK

TUTORIAL

How to edit images on ChatGPT/DALL•E:

Once you’ve had DALL•E generate an image, click on the image itself to make it full screen, at which point you will see the “edit” button on the bottom right, underneath the generation. It will allow you to type desired edits on both mobile and desktop, but on the browser version you can select regions to micro-edit, and can keep editing until you achieve the desired results.

This is a great way to create slides, infographics, and still images to animate using a text-to-video platform like Runway ML, Pika, etc. (both featured in previous issues of AI Logs).

TOOLS OF THE WEEK

🗣️ Lazy is an AI app builder that’s surprisingly easy and delivers pretty quality, although fairly basic, apps. Free to play with, if you generate an app you actually like and want to promote and potentially even monetize, the cost is beyond reasonable at around $40/month with small usage costs that scale up. I played with this to spin out several types of apps already, and am genuinely impressed.

🤖 OpenUI is an open-sourced W3C generative AI for UI (user interface; front-end dev) that is a great compliment to the many g-AI development tools, many of which I’ve featured here previously such as Open Devin and Devika. With the capability of both front- and back-end development using AI, the software game and its capability of turning ideas into digital reality is definitely redefining itself.

🧐 Upmetrics is a great AI that can generate pitch decks, business plans, financial models, and much more. Highly editable and decent at market research and strategy, this tool can help non-experienced founders “fake it until they make it”. It can also help seasoned founders save tons of time and get to market (or investable) much quicker.

🎲 LangFlow for LangChain is an AI to create and share AI research, workflows, autonomous agents and much more. Just acquired by DataStax, a RAG API and Database-as-a-Servixe, LangFlow could help level the playing field for many younger AI startups and help many leverage bleeding edge technology in their current ventures. Not for the purely non-technical AI users yet, this Python based tool is definitely one everyone should at least be aware of, and that could scale into something most impactful.