✨#howtoai ✨

Let’s go @nyknicks #howtoai

Wise wasp is the Yoda. (@StuffKeithBuys ) The cat is the Pikachu (Julia) Who inspires you? Let’s go @StuffKeithBuys Thanks for helping me learn #howtoai Fine Kow!

@grok another banger. #howtoai

Generated with @grok imagine #howtoai #garyvee #foreverknicks

i dont expect "howtoai" to have a functioning brain

This Viral Post by HowToAI is a Shocking Indictment of the State of AI discourse on X This myth of a Single AI “Brain” in this viral post must be debunked, for the bullsh1t it is, bullsh1t that gets boosted on X while facts, the falsifiable, and the truth get throttled. The viral post by HowToAI perfectly demonstrates the myths of AI and the contemporary AI-hype theatrics of low-informed and organised AI boosters. This post sensationalises an MIT-affiliated paper as if it proved that all big AI models share one “brain.” In truth, the Platonic Representation Hypothesis paper is a speculative survey, not a proof. The authors themselves frame it as a hypothesis and explicitly note many caveats. The abstract says “we argue that representations… are converging” and “we hypothesise a shared statistical model of reality”, language that clearly stops short of a definitive claim. In fact, the paper ends by highlighting counterexamples and open questions, admitting that their findings are suggestive and that “different models arrive at similar but not the same representations”. In short, the work is interesting food for thought, but it does not prove the existence of a universal AI brain. Convergence Metrics ≠ Identical Geometry The author of the post claims “the mathematical geometry is identical” across modalities. This is a severe overstatement. The paper measures alignment via similarity kernels (e.g., mutual nearest-neighbour overlap), rather than the exact identity of representations. They find that as models scale up, alignment tends to increase, but remains far from perfect. For example, the best reported alignment score is only about 0.6 on their metric (where 1.0 indicates perfect overlap). The authors stress this point: different models “arrive at similar but not the same representations”, and a score of 0.6 (out of 1.0) leaves a lot unexplained. A follow-up study by Gröger et al. (2026) further cautions that popular alignment metrics can be misleading: after statistically correcting for trivial effects of model size, “the apparent convergence reported by global measures largely disappears”. In other words, the partial similarities observed are mostly low-level trends or shared biases, not evidence of a single unified “geometry.” It is simply false to say vision-only and language-only nets measure distances exactly the same way. They have overlapping structure, but also big differences. Modalities and Data Matter Another false claim is that “it doesn’t matter what data” or architecture, as if any large model is forced to map the one “true reality.” In reality, different data sources carry different information. The authors explicitly acknowledge this: “two different models cannot converge to the same representation if they have access to fundamentally different information”. For example, a text model never directly sees colours or shapes, and an image model has no native notion of abstract concepts expressed only in language. The paper points out that their convergence argument strictly holds only when the input projections are bijective (i.e., information-preserving), a condition rarely met in practice. In fact, the authors show that richer (more informative) captions improve vision-language alignment, implying that sparse data limits the effectiveness of any common representation. In short, there is not “only one reality to map” for AI: each model’s “view” of the world is filtered by its data and tasks. The paper even warns that highly specialised models (say, one trained just to drive a car or predict proteins) may follow “shortcuts” and not share representations at all. So the posts claim that any large model inevitably builds a universal world model is unsupported. Interpret with Caution In summary, the posts claim wildly overreaches. 1. The Platonic Representation Hypothesis is a theory, and a cautious one: it suggests some representational similarities may emerge under certain conditions, not that AI models have identical “brains.” 2. All the paper’s concrete results show is a trend toward more alignment with size and multimodal training, with many caveats. 3. The authors themselves end by calling the remaining differences an open question. 4. A recent re-analysis confirms that most of the apparent convergence vanishes under stricter measurement. In short, AI models do not measure distances between concepts in exactly the same way, nor do they discover a single Platonic structure. The hype ignores the limitations sections of the paper, cherry-picks language like “shared model of reality,” and ignores that even the authors say “different sensors… might capture different information, which may limit their potential to converge”. So, contrary to the post's claim, no one has proven the existence of a universal AI brain, and the Platonic hypothesis remains an intriguing idea, rather than a demonstrated fact. Sources: The arXiv paper by Huh et al. (2024) discusses these trends and caveats. A follow-up study by Gröger et al. (2026) critically examines the metrics and finds much weaker evidence for convergence. These are the actual sources behind the hype. Huh et al., “The Platonic Representation Hypothesis” (arXiv:2405.07987, 2024); Gröger et al., “Revisiting the Platonic Representation Hypothesis: An Aristotelian View” (arXiv:2602.14486, 2026).

Want to learn AI/ML but don’t know where to start? These YouTube channels will guide you step by step. Whether you're building Agentic AI workflows or just starting with Python, this llist covers everything: 1. Agentic AI & ChatGPT Watch creators like Krish Naik and Freecodecamp for hands-on projects with GPT, LLMs, and automation tools. 2. Machine Learning & Deep Learning Explore channels like StatQuest, Yannic Kilcher, and DeepLearningAI for algorithm intuition and implementation walkthroughs. 3. Python & Data Science From Corey Schafer to Ken Jee, these channels help you master the foundations of Python, analytics, and deployment. 4. Math for AI Need help with the math behind the models? Start with 3Blue1Brown, Khan Academy, and PatrickJMT. 5. Big Data & Data Engineering Follow Alex The Analyst, edureka!, and Data Engineering to scale your AI pipelines and manage large datasets. 6. Tools, Productivity & AI Assistants Stay up to date with Ansh Mehra, Jeff Su, Skill Leap AI, and Howtoai to explore ChatGPT hacks, AI agents, and productivity automation. Start with 1–2 channels that match your current level and niche. Be consistent, practice what you watch, and explore new tools as you grow.

BOOK ALERT! There are lots of books out there on #AI, which is the latest buzzword. But if you are looking at a primer that gets you up to speed with AI, how it works, its major players and how you can make the most of it, we recommend the very aptly named "How to AI" by Christoper @mims, available in India from @HachetteIndia . It is super easy and even fun to read, and at around 250 pages, not an intimidating read at all. @NimishAndAkriti review it in @IndianExpress, in our #ReadIt column, calling it "perhaps the one artificial intelligence primer all of us need to read." #AI #HowToAI #ChristopherMims #Book #BookReview #Reading indianexpress.com/article/bo…

just the howtoai youtube channel ALONE was making me $100k/mo at age twenty but the reason why I didn't bring that operation to $150K-$200K/mo was not due to what you might think the number #1 killer of scaling is getting stuck in the day-to-day trenches of reviewing scripts, managing freelancers if I could start one or two howtoai copies at that time, I would, but my time was too restricted this is a costly lesson I've learnt in hindsight and also the reason my own software was born this bottleneck of being stuck in the grunt work affects 99% of content agencies, YTA creators & marketers many entrepreneurs could easily 2x their income by dedicating a week to making hires, writing SOPs & structuring up their team management system if this sounds like you, sign up to the beta wailist on the @kloudboard page also send a DM with your email for priority acceptance it will not only save you money on subscriptions, but most importantly earn you a lot by helping you scale super exciting times ahead

now can you still do all of this and get demonetized? YES that's just the game I had a channel called howtoai that got terminated in 2025 I wasn't posting AI slop, my content wasn't violating youtube's rules and it still got terminated that's just the way it is sometimes but that doesn’t mean you shouldn’t start youtube automation it’s still one of the best business models you can start as a beginner you just need to approach it the right way: 1) create one channel 2) learn the process 3) build your team (people AI) 4) scale that channel to $5k /mo 5) then start a second channel 6) then a third one don't put all your eggs in one basket have channels in different niches so if one gets hit, you're still safe that's how you build a sustainable youtube business

Just 9 videos? 100 QUALITY videos that people watch 100% before the algorithm likes your channel. But now YT cracks down on AI slop and delete AI channels, see the channel HowtoAI?

Want to learn AI/ML but don’t know where to start? These YouTube channels will guide you step by step. Whether you're building Agentic AI workflows or just starting with Python, this llist covers everything: 1. Agentic AI & ChatGPT Watch creators like Krish Naik and Freecodecamp for hands-on projects with GPT, LLMs, and automation tools. 2. Machine Learning & Deep Learning Explore channels like StatQuest, Yannic Kilcher, and DeepLearningAI for algorithm intuition and implementation walkthroughs. 3. Python & Data Science From Corey Schafer to Ken Jee, these channels help you master the foundations of Python, analytics, and deployment. 4. Math for AI Need help with the math behind the models? Start with 3Blue1Brown, Khan Academy, and PatrickJMT. 5. Big Data & Data Engineering Follow Alex The Analyst, edureka!, and Data Engineering to scale your AI pipelines and manage large datasets. 6. Tools, Productivity & AI Assistants Stay up to date with Ansh Mehra, Jeff Su, Skill Leap AI, and Howtoai to explore ChatGPT hacks, AI agents, and productivity automation. Start with 1–2 channels that match your current level and niche. Be consistent, practice what you watch, and explore new tools as you grow.

update: @TeamYouTube seems to have completely deleted all data of the channel from the platform the channel's entire existence has been removed from the platform, including any data they were storing this seems to be a new update, as the channel would remain with its data, but in the terminated state now though, they seem to be getting rid of everything might be wrong, so let me know if you have any insight on this was a good run with howtoai, made a big mistake not saving all my content

this is why howtoai was insanely profitable it made 2-3x the amount of money a 7M subscriber personal brand does even when it was only at 100k subscribers got a store at $1.5M revenue for a single product I was promoting, and I was monetizing through many more either you go the ad revenue route and go infinitely hard on volume and scale, or niche down and provide value with your content allowing for external monetization with your own products / affiliate deals either works, and people are making insane money from both, but I see too many beginners trying to do both there is no right or wrong in this question, but if you're trying to sell an AI e-book to your entertainment channel audience, chances are it wont work efficiently with a channel providing educational content the viewers are already buyers the second they click your content you already know their intent by them just watching the video then just weave in your product to the solution and you're done gg