GPT4"lite"; a disruptive prompt, competition escalates (+ win $1,000) TGF005

If the past few weeks or even months have seemed surprisingly slow relative to AI news, the same will likely not still ring true this time next week. In today's issue, I'll go over a little house keeping, will discuss Visualization of Thought (VoT) prompting and implications, will unpack several OpenAI rumors and their 'race' with 谷歌 (and now somehow their $13B+ investing partner 微软 ), and encourage anyone interesting in how to make music and video using AI to participate in a content I've launched on and with BlockBook . Let's get into it!

Knock, knock: "Housekeeping"...

"Housekeeping who?"

"Housekeeping up with all the current AI trends treating your hairline, Cory?" ??

But (bad) jokes aside, to start off with some housekeeping, my previous AI newsletter, "AI Logs", is still in production, just no longer with me as editor-in-chief, and although they still closely resemble my topical subject matter here, I am completely removed from that project - if you're still subscribed to it, please simply know that it is no longer part of my portfolio.

Secondly, my sincere gratitude to each and every one of YOU who's subscribed to this 'Tech For Good: Building In Public Using AI' newsletter - over 120,000 and it's still the first month of publication. I'm blown away to see builders and decision makers from companies such as the ones tagged above subscribing to this - your interest, confidence, support and feedback does not go unnoticed.

Thirdly, I have named this 'building in public using AI' for a reason - namely, I want to empower YOU with the knowledge of how to use available AI tools like a PRO (they're their own user manuals, by the way), see how I, my teams/ventures and others are using different AIs to code, build, test, grow, connect, research, prototype, design and otherwise make impact bringing ideas to life and market, and to help prepare as many of us as I can for the artificial general intelligence (AGI - what some would refer to as "real artificial intelligence) that's coming VERY SOON and how to get AI to actually do what it is that we want it to, which often comes down to mastery of prompting (sometimes referred to as 'prompt engineering') - the significance of which I will get into next as I dive deep into the concept of Visualization of Thought (VoT).

Visualization of Thought prompting is the next-level from Chain of Thought (CoT) prompting, which I've discussed in previous issues. As a quick reminder, CoT is a prompt engineering technique that helps large language models (LLMs) improve their reasoning abilities by having them explain their reasoning each step of the way to reach it's 'final answer'. Meta AI was recently unleashed upon billions of users of Meta products Facebook, Instagram, WhatsApp and Threads, and their Head of AI (Chief AI Scientist) was recently quoted as saying we (society) can't/won't achieve AGI "with large language models alone", as they're incapable of reasoning, understanding 'how things work', etc.

I have written and posted several times about the correctness of this, but also speculated that other models such as large world models (LWMs, most notably popularized by OpenAI's Sora), large nature models (LNMs) and large action models (such as Rabbit R1, which has now been replicated as a simple Android app that doesn't require their proprietary hardware device: file under "things that make you go hmmm") may collectively get us to - or at least incredibly close to - AGI (which is often considered to be AI able to do anything a human can at a similar level of competency, and allegedly will reach an IQ of well above 1,200, or more than 9 times that of Einstein's).

Well, this new style of prompting and using LLMs of VoT may be a huge stride forward in the quest to achieve AGI (OpenAI's new declared mission statement as well as Mark Zuckerberg's), as it "resembles the mind's eye process" by having multimodal LLMs visualize their reasoning traces, effectively 'showing their math' visually. Learn more about VoT and the implications in the Microsoft-sponsored white paper here https://arxiv.org/pdf/2404.03622.

This really goes to show how figuring out superior ways to "talk to" the AI tools and prompt them for superior results can really be game-changing. It reminds me of months ago when I got ChatGPT to start prompting itself briefly, taking me totally out of the loop (that got shut down real quick!), and foreshadows what we can expect soon with agentic systems, or autonomous agents, which is likely going to be seen in coming ChatGPT models as well as Microsoft AI and Google's.

Now to address some ChatGPT and OpenAI speculation - last issue I wrote about "chatgpt-chatbot" which was a spontaneous, open-sources, mysterious new LLM/AI available on LMSYS Arena said to outperform even ChatGPT4 Turbo, with Sam Altman (SamA) trolling it, mentioning it, and then denying at Harvard that it is really "ChatGPT5". However he did NOT say that it wasn't ChatGPT4.5, nor mention of "ChatGPT4 Lite" which has been referenced internally at OpenAI for some time. He (Sam) also said that 'ChatGPT5 will be way better then GPT4, in general'. I think the word "general' may be very telling, as Sam is a very literal person, and general is the G in AGI (again, 'artificial general intelligence').

Further OpenAI 'news' is that they'd been planning a big announcement this week, but have moved it to this upcoming Monday, the same day as Google's big IO developer conference, where they'll no doubt make some big AI announcements of their own; that said, OpenAI loves to overshadow Google's AI announcements, so the timing of moving their (OpenAI's) announcement from this week to next week seems potentially quite revealing, particularly as it's been revealed that OpenAI is planning to launch their own search engine, competitive with Google, Microsoft's Bing, and AI darling, Perplexity .

Although it may not be next week, we can - with confidence - anticipate a multimodal version of ChatGPT soon with access to Sora text-to-video (LWM), autonomous agents (the AI can prompt itself and act on the users' behalf), active reasoning, access to it's own search engine ('the web') and more. The time to learn to prompt, understand and be understood by AI tools, and get them to do what we need and want them to do for us is upon us - I'm an optimist, but it's really the time to "keep up, or be left behind". The fact that you've made it this far into this issue shows that you're in the "keep up" camp with me - I applaud us both.

So while we don't know if "chatgpt2-chatbot" or the "new versions" codename 'i-am-a-good-gpt2-chatbot' etc is actually made by ChatGPT (it says it is), and don't know if it's Q*, GPT5, AGI or something totally different, we do know that it's on SamA's radar, that is performs great, that it is ahead of a big OpenAI announcement/release, and that it has the AI community buzzing.

Lastly (just to keep it somewhat brief - I'll be publishing these twice weekly for now as there's so much to cover and never enough time/space) I have been teaching how to make music and videos using AI for free for some time, and also used AI to create "the worst" rap song and "worst music video" consistent with my joke that "AI is getting a bad rap" (so I created it for it). For those who want to learn, and want to showcase their skills making a GOOD (not bad - this newsletter should only ever be TECH FOR GOOD, lol), I encourage you to check out my spark on Blockbook where your submission could win the following prizes:

??1st place ($10,000+ value):- $1,000 USD CASH - Guaranteed 1M views with prominent mention on social media (Priceless)- 1 year WeJ Premium ($2,364 retail)- A mention in top global AI newsletter ($5,000 retail)- File under “PR”, a branding session ($1,250 retail)- Viral growth masterclass ($750 retail)

??2nd place ($2,300+ value):- $300 USD CASH- File under “PR”, a branding session ($1,250 retail)- Viral growth masterclass ($750 retail)

??3rd place ($850 value)- $100 USD CASH - Viral growth masterclass ($750 retail)Join us in this contest ("spark") to earn massive exposure and exciting prizes!

Enter the 'Spark' and learn more here ?? https://lnkd.in/ej39pkeH.

Basically, there are links there ^^ to see my awful AI rap "Oops I Dropped A Flop" and to learn to make your own music and videos, and if you create one you want to submit, it's possible via the link above. I love what Blockbook is launching for influencers/creators like me, and have been loving the submissions to this contest already as well.

I said lastly, so will honor that, but in closing I'd love to encourage you to join the hundreds who've already provided me with feedback, suggestions, and critique of this newsletter. My previous AI newsletter, AI Logs, had weekly tutorials and prompts - if you'd like me to bring those back, or other such resources, please let me know. I am also open to making these more visual or collaborative, if the appetitite is there, so definitely let me know (I know they're quite text-based here yet, in contrast to my previous one). THANK YOU again for being a subscriber here - I'll likely launch a companion newsletter on beehiiv soon, as well as a private community (status quo is discord, but am entertaining WhatsApp or Slack as well) for those that crave EVEN MORE. The fact that you've read this entire issue - issue #5 - says a lot about YOU. I appreciate the intention, interest, willingness, adoption and participation. As I say here often: TOGETHER WE RISE. Thanks for being on this journey with me. Signing off,

Cory Warfield