A.I. wants to be the worst employee you've ever had; Microsoft breaking up with OpenAI; "DeepSeek moments" aren't a thing - Tech For Good #TFG025

Ma, Ma, Mai. First off, I'll unpack this title so you know if this issue will be worth the 8 min read, or if you should just catch the TDLR or summary from your LLM of choice (or other favorite influencers' summation), but then I do need to revel in a recent win (today) regarding this newsletter that is an absolute affirmation point that's been made possible by you, my amazing Tech For Good community of over 200,000 subscribers after only 24 issues (but that's not the win I'm referencing).

Ma, ma, mia... is it a typo, and if not, what do I mean by it? Well, recently Manus, an A.I. agents platform by Chinese startup 'Monica' has recently taken headlines, as has Manas AI , and Microsoft AI is reportedly distancing itself from OpenAI in many ways, including building their own LLM that is not ChatGPT to potentially power Copilot and the 微软 ecosystem, including LinkedIn .

Also, probably most important in my assessment, quite candidly, is the new Diffusion LLM from a startup called Mercury from Inception Labs , which is the technology that actually prompted me to do a new issue of Tech For Good, because I truly think it's that disruptive and "good". So if you want to learn about, geek out about, or critique my understand of agentic frameworks and where we're at today with them, a totally new methodology of coming up with awesome results using A.I. for up to 10 times faster and 10 times cheaper but also soon 10 times better, where we're at societally with A.I. and A.I. adoption (and how to double DAUs to reach half a billion humans daily who are consciously using and fully able to leverage A.I. - which I propose is actually just intelligence or advanced intelligence, rather than "artificial"), then keep reading - otherwise see what your favorite YouTuber says about it, pop the URL into your LLM of choice for a summary, or just catch it next week. THANK YOU! I truly appreciate all the support, feedback, encouragement, messages, comments and momentum.

To that end, I woke up this morning to a message from a founding executive at an A.I. company I've reviewed and mentioned here previously, and who've raised a significant amount of capital recently and who I actually use on a somewhat frequent basis. They are interested in sponsoring or being featured in this newsletter and asked for my media kit, they said they've been fans of this newsletter since it's earliest editions, and referenced a use-case/tutorial from months ago. Obviously, we're all in business in some form or fashion, so inbound interest for anything I'm doing whether my coaching, consulting, influencer marketing, investment into any of my numerous ventures etc. is always welcome and affirming, but knowing that my perspectives, my research, the ways I used and also potential frustrations I've perceived with emergent technologies - maybe even theirs - motivates me to keep at this, tells me that "I'm on to something", but also can affirm that if I can do it, (almost) anyone can. Sure, this is now one of the largest and fastest growing A.I. newsletters in the world now, and I have a major platform here on LinkedIn now with 580,000 followers etc., but all of this has been the byproduct of me just continuing showing up, sharing my thoughts while holding space for you al to share yours, and putting myself out there.

And special shoutout to XXXXX-XXXXXXX and XXX, who will likely be featured in upcoming issues and posts of mine (but I will not artificially inflate, provide biased reviews or recommendations, nor is TFG a "pay to play" publication).

So you've got the agenda, an intro that hopefully inspires some of you, and now let's get into it - let's check out Act One of:

Ma, Ma, Mai --------------------------------------------->

As a music head, this makes me think about "My Sharona" by The Knack, Skeelo "I Wish", and even Hammer fka MC Hammer "Can't Touch This" (ma, mai music hits me so hard, makes me so: oh my Lord; Thank You for blessing me...), but on a different level, I think

• Manus
• Manas
• Microsoft AI (MAI), and
• Mercury Diffusion LLMs

will all be as big of game-changers as test-time compute, process-based rewards, "more compute" and other ways to level-up what A.I. is capable of - much of which we've been and remain solving for humanity at LovingIs... A.I. <<< open in limited open-alpha; please follow our page to learn of upcoming big advancements and announcements.

MANUS

So let's start with the "elephant in the room" of the week (replacing DeepSeek as the flavor of the week) which is Manus, by Monica - an agentic system that's quite capable in performing complex tasks, deploying a mixture of experts, completing deliverables such as presentations, games, research projects and more very well at this stage, and which has may other influencers in a frenzy with "how good it really is". To be clear, I have not personally played with it due to some personal concerns and hesitations, but I've spoken to many contemporaries who have, have watched hours of others' usage of it live/in real time, and I will say, with my limited working knowledge of it, that it does seem quite good - but shouldn't surprise anyone who's been fully dialed in and connecting the dots of what's coming. It's agentic, yes; it can do tasks for us, identify what to do and how to do it, and do a lot of what many businesses and business leaders have delegated to humans for generations. That said, it doesn't seem to be much smarter than Claude 3.7 (which might power it), ChatGPT 4.5 (now live for most to try, many are underwhelmed), Grok 3 (I've reluctantly admitted to being blown away by this model, personally) - if smarter at all - it is great at working semi-autonomously, knowing how to delegate, prompt, review deliverables, and more. Once this can fully take over our terminals and devices, and act on our behalf wherever we humans are currently required to be authenticated, etc., it will unlock a lot for potential good, but also for potential bad.

MANAS

I've mentioned it here and in my daily Linkedin content, but recently Reid Hoffman launched Manas AI as a preeminent leader in the usage of A.I. to both detect but also prevent and treat cancer, as well as many other diseases and illnesses that have plagued humanity for its existence. While many other companies including Google DeepMind , 西门子 Healthineers and more are also contibuting significantly, as well as visionaries such as Abby Aboitiz ( AI Wellness ), Dr Nik ( theHotBleep ), Fred Amaya ( Nudora Probiotics ), Ai Phi Thuy Ho, MD ( NorVue ), Sharon Kedar, CFA ( Northpond Ventures ), John Golden ( Vital Neuro ), as an adult who's lost both parents to cancer, I am a huge advocate for what Manas AI is doing, and will not cease to continue highlighting their powerful work, significant investment into and relentless fight against cancer and other humanity-limiting ailments.

Interestingly, it's when we see companies like many listed above and Agora World , Verascan Inc. (soon to be KENEK) all working together, especially with ways to scan the body from both the outside (POCUS) and inside (nanobots) collaborating with each other that we may see the human life expectancy increase by several orders of magnitude.

MICROSOFT AI (MAI)

And now, before our "MAIN STORY" of Diffusion LLMs and why they matter is the matter of Microsoft's new AI models - MAI, and why they matter. Well, after investing a whopping $13B into OpenAi for almost 50% ownership, a clause was revealed that stated that when and if OpenAI researched AGI (general artificial intelligence) that Microsoft would lose it's ownership (some wonder if they already have, or if the DeepSeek approach has MS believing they can overcome OAI technically very quickly with "more compute"). Microsoft then spent over half a billion dollars effectively buying its board member and founder of it's property, LinkedIn, the aforementioned Reid Hoffman, 's A.I. upstart, Inflection AI , days later licensing its voice model to Google, then OpenAI. To have invested $13B, then an additional $650M, and now whatever they're spending outside of their partnership with OpenAI on MAI is very telling.

I'm not sure what it means, nor do I believe many, if any, truly do - but I do think it may mean we're closer or further from "AGI" (closer would make way more sense), that people "at the top" "know something" "the rest of us" don't, and that we're highly likely to see yet another highly capable model released to the public soon. Once agents fully roll-out

WHICH OPEN AI IS ALREADY DOING AT SCALE

then I think these models, especially those who figure out multimodality and advanced reasoning, will be capable of a lot. Especially those with

DIFFUSION LLMS

like the one we saw released this past week by Mercury.

Why does the term 'diffusion' already sound so familiar in the A.I. vernacular? That would be from Stable Diffusion from Stability AI , which was an early disruptive text-to-image generator that also uses the diffusion methodology.

How does it work and how is it different?

Instead of sequentially trying to create an image pixel by pixel, or token by token, Stability took the opposite approach, quickly generating a full, imperfect (think 'blurry') rendering and then quickly adding to it, refining it, and making it better and better, until finally providing the generation to the user. This is precisely what the diffusion LLMS can produce with tokens rather than pixels - in other words it can generate an entire book, script, report, presentation etc. as an abstract or 1% of the final deliverable, and fine tune it from there.

So far this methodology is proving to be around ten times faster, ten times cheaper, and more capable than the existing sequential generation that other LLMs employ.

So there you have it - the huge leap in intelligence within intelligence that can help us all be more intelligent... or something like that. To those who've been requesting demos, I'll be doing a live one this weekend with my friend Adam Gaskill of The Nonprofit Creators , and feature some throughout my weekly content on my personal Linkedin page.

To those with questions, insights, thoughts, comments, concerns - please leave them in the comments below: I do check them all, as do many of my readers. Thanks to you all and your support Tech For Good newsletter is closing in on 220K subscribers (nearly 10K per issue!) and my #techforgood content is reaching roughly 2M impressions weekly and my profile is closing in on 600K followers. I am still selectively onboarding new viral growth coaching clients who want to experience similar reach and growth who have a message, technology, heart and alignment to my ecosystem - feel free to reach out to learn more. I'm also available globally as a keynote speaker via PepTalk .

And now, as I like to do, let's finish with an A.I. generated cartoon - please submit any of your creations to Techitoons .

As a point of clarification, I'm not trying to poke fun at or shame Reid here - but I wanted a cartoon of a tech founder who's avatar wasn't amused by the human's humor, and specifically did not want it to represent Mr. Hoffman, and the fact that this character has the belt buckle, the wall hanging, and the blatant typos in the 'punchline' show how far we really are from wanting to entrust our actual work deliverables to A.I.s - even OpenAI's aspirational $20,000 / mo. agents.

If the A.I. can't spell, follow simply commands, get biology right like fingers, feet, etc., and physics like reflections, shadows etc. right, and "hallucinates" et al., I think we humans can still expect to have some job security for the short term... that said, we should all be thinking about UBI (universal basic income) and other ways to stay ahead of the impending inevitable.

Let's tread this course together, and lightly. And now, let's diffuse...

(Also, PRO TIP: feel free to peruse previous issues of Tech For Good as they're still highly relevant and ahead of the curve, and also feel free to share with friends - sharing is caring!)