Valentin Matresu

I'm passionate about building with AI, finance/crypto, tech and personal development. More than 20 years in software development.

[Brian Roemmele @BrianRoemmele]

WOW! The $8 AI Machine!

Something extraordinary just happened and it changes what “local AI” can mean.

I am testing it tonight. Thus far it shows many possibilities…

So what it this $8 AI device?

A developer going by slvDev has forced a 28.9-million-parameter language model onto an ESP32-S3 microcontroller that costs roughly eight dollars.

Not a Raspberry Pi.

Not a Jetson.

An eight-dollar microcontroller.

The model runs completely offline, generates coherent short stories at about 9.5 tokens per second, and draws power measured in the same range as a small LED.

This is more than a hundred times larger than the previous record for the same class of chip (the earlier 260,000-parameter TinyStories experiments).

For perspective, the original ChatGPT sat at 117 million parameters. We are now running a model roughly a quarter of that size on silicon you can buy for the price of two coffees.

How the Impossible Became Possible

The ESP32-S3 has only 512 KB of fast SRAM, 8 MB of PSRAM, and 16 MB of flash. Conventional wisdom said a model of this size simply would not fit.

The breakthrough is architectural, not brute force.

Most of a language model’s parameters live in a giant embedding table a lookup table you mostly read from, not compute against.

Drawing directly from Google’s Per-Layer Embeddings technique (the same family of ideas used in the Gemma models), the developer moved the bulk of that table roughly 25 million parameters into flash memory and memory-mapped it.

The chip only needs to pull about six rows, roughly 450 bytes, for each new token. The remaining dense “thinking” core stays in the fast SRAM (around 560 K of active working memory). The model is stored at 4-bit quantization and occupies about 14.9 MB total.

The result is a system that feels almost free to run. The heavy parameters sit quietly in flash and are sampled sparingly. The little core does the real work. It is elegant engineering of the purest kind.

What I Am Doing With It Right Now

I have the boards on the bench in the garage lab. The first units are already talking short, coherent stories appearing on a tiny wired display, generated entirely on the chip with no Wi-Fi, no API key, no cloud round-trip. Latency is local. Privacy is absolute. Power draw is low enough that battery operation becomes interesting.

I am treating these as the first generation of true $8 AI machines. Early tests are focused on three practical directions.

- Embedding the model into simple nodes.

- Pairing it with local voice front-ends

- Exploring whether multiple of these chips can be networked as a lightweight swarm.

The model is deliberately limited. It was trained on the Microsoft TinyStories dataset and is excellent at coherent narrative, not at open-ended question answering or tool use.

That is a feature, not a bug. It forces us to design systems around what the silicon can actually deliver instead of pretending every edge device needs a frontier model.

Real Use Cases That Suddenly Become Practical

Once you accept that a capable language model can live for eight dollars and run without the cloud, a new class of devices becomes possible:

This is the opposite of the current trajectory that wants every intelligent act to travel through a remote server. It is the beginning of intelligence that is cheap enough, private enough, and local enough to become infrastructure rather than a service.

We have spent years watching model sizes explode upward. The more interesting frontier may be the opposite direction: how small, how cheap, and how local can useful intelligence become? An eight-dollar chip that can tell coherent stories is not a toy. It is a proof that the lower bound keeps moving.

The open repository is at https://t.co/a7gcHTR4ug

I will keep testing, measuring, and reporting what these little machines can and cannot do. The age of abundant local intelligence just got a little more real, and it arrived wearing an eight-dollar price tag.

[iPaul @iPaulCanada]

2026世界人工智能大会，某科技公司展示的跨时代教具--AI黑板

[aditya @adxtyahq]

KIMI K3 IS A NIGHTMARE FOR CLAUDE FABLE-5 AND GPT-5.6-SOL FOR GAME DEVELOPMENT

Tried building a playable MMORPG with Kimi K3 featuring horse riding, rooftop parkour, combat and an explorable open world.

The whole thing took about 55 minutes from a single prompt and cost exactly $9.37 to generate.

[Hari @hrkrshnn]

I asked the new Kimi K3 to pentest a website and looked into its inner reasoning traces.

It has references to Anthropic's usage policy! Kimi clearly distilled Claude's cyber capabilities.

[Kfir Aberman @AbermanKfir]

1/ Today, we’re excited to introduce Lucy 2.5.

Lucy represents a paradigm shift in world models, not just in how generated worlds look, but in how we interact with them.

A thread on what makes it a paradigm shift 👇

[Burcun Ne? @NeBurcusun]

Fizik tedavi hocam bana bunu söyledi ve bunun öne eğik boynu (text neck / turtle neck) düzeltmek için en iyi yöntem olduğunu söyledi.

Bir hafta boyunca uyguladım ve boynumdaki tüm sertlik tamamen kayboldu. Hatta o kadar alışmışım ki, sertlik gidince garip hissettim.

Her gün 10 dakika boyunca bu şekilde yüzüstü uzanmayı deneyin.

Bunu yaptığınızda ensenizdeki o gerginlik ve sertlik hissi kayboluyor. 😱

[enzo @enzo_gte]

I've been using Kimi K3 for ~16 hours now.

The model is clearly good at a lot of different things (especially frontend), but non obvious reason why people are enjoying it so much is that it clearly does not follow the same rules in terms of safeguards and copyright.

Kimi will happily clone MacOSX. If you ask it to help you improve another AI model, it will do it with a smile on its virtual face.

Ask Fable to do the same thing? It literally starts to perceive you as a criminal committing a war crime (like no bro, all I want to do is fine tune an open source model).

After using all three recent releases, Fable, GPT 5.6, and now Kimi, it's clear that the full power of the models has been significantly held back by the safeguard restrictions caused by last months debacle with the USG -- leading to the top models being quite literally lobotomized in some areas, which leads to subpar results as the safeguards pollute its entire thinking and problem solving abilities.

The funny part? Is that you could have predicted this outcome 2-3 years ago when you started to see the rise of Chinese EVs and smartphones compared to western alternatives.

They quite literally tried to copy the Tesla Model S and iPhone as hard as possible and then eventually it started to diverge to the point where their EVs and phones are just genuinely better (which is why we have export controls banning their EVs, because they would literally drive all US manufacturers to ZERO)

There is a very clear behavior difference in Chinese capitalism and American capitalism.

American capitalism tries to protects copyright, patents, etc (oh no, you can't download a book through LibGen, that's ILLEGAL!).

Versus Chinese capitalism actually just does not give a fuck.

"Hey you want a video gen model (Seeddance 2.5) trained on every single anime ever? And you want the main character to look exactly like Messi? Sure, here you go!"

You see what I mean? When one half of the competition is being held up by regulators and restrictions on people who don't understand the technology and the other half has a leader who quite literally today said they are going to set up AI centers around the world to help other countries onboard to their open-source AIs, this is the sort of results that you will start to get.

These models were not smart enough to have this difference in philosophy matter -- but the newest class of models is where this difference makes a big deal. If these models are finally at the point where they are smarter than 99% of humans, why would you want to use the American one who tries to impose its world view onto you versus the Chinese one who will just do what you say without asking any questions?

And this isn't a full on bullpost on Kimi, the model is clearly not as smart as Fable / GPT 5.6 on things like math and science, but it's lack of handcuffs means that it can show the world what the frontier labs are gatekeeping from you and that starts to build customer resentment and loyalty towards the East, which is probably not what the USG wants.

Interesting times. Interesting times, indeed.

[THE BITCOIN PHARAOH @thebtcpharaoh]

On Saylor

In my twenty years of investing I've been in rooms with a lot of highly intelligent people. I've also followed and carefully studied many others. When I first came across Michael Saylor (on Lex Fridman's podcast a few years ago) it was immediately clear to me that he was perhaps one of only a handful of the most visionary people I've ever come across.

The company he keeps in my head is small. Steve Jobs, who understood what what people wanted before they did. Jeff Bezos, who held a single conviction through years of public ridicule. And perhaps a couple of others. Saylor is an MIT engineer running a balance sheet like an instrument, in service of a conviction the market still ridicules. He belongs in that room.

So I want to be clear about the piece below, and give it the frame it deserves. When I write my ideas and brainstorm openly, I'm second-guessing nothing. I have no doubt that Saylor is actively considering every option available to him, including the ones it takes me a couple of hours (with no consequence) to write and post. He ran the same math long before I did, with better information, tools, experience and IQ that trump mine on any given day. There’s a reason he is where he is and we are where we are.

I write pieces like this because I enjoy it. I like to articulate my thoughts and debate ideas openly that challenge my own thesis and help me in my investment decisions.

I don’t study the pyramids because I think I could have built them better.

[Valentin Matresu @valentinmatresu]

@saylor @Strategy and from “Everybody in the world would love to have a high yield bank account that yielded 10% or more” / pitching STRC like a ~12% bank-account-style product, to treating “keeping STRC at par” as no longer important for Strategy?

[Valentin Matresu @valentinmatresu]

@saylor @Strategy For what? So you can flip-flop from “You do not sell your Bitcoin” / “Never sell your Bitcoin” to “I said you shouldn’t sell your Bitcoin. I never said the company wouldn’t sell some of its Bitcoin”),

[Valentin Matresu @valentinmatresu]

@BitStrategy21 "Never sell your Bitcoin" oh, wait I meant you, not my multi billion dollar company. "We will always be buying at the top" and now selling stock and Bitcoin at the bottom to fund USD cash reserve. And many others that were switched around.

[BitStrategy @BitStrategy21]

When $STRC initially fell away from par, many analysts suggested it would quickly revert, as they believed traders would step in to 'correct' the market.

While some traders did step in, including Strategy CEO Phong Le, it was insufficient to counteract the downside momentum in the short term.

Now the rapid reversion to par has not occurred, these same analysts claim that STRC price is primarily driven by Bitcoin price and sentiment.

To maintain intellectual honesty, it is important for us to acknowledge when our prior reasoning was incorrect.

[Thinking Machines @thinkymachines]

Today, we are introducing Inkling.

Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.

https://t.co/Ghebq5mG30

Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵

[Xenova @xenovacom]

Bonsai 27B just changed the local LLM game forever.

1-bit quantization shrinks it from 54GB to just 3.8GB (-93%), while retaining 90% of its intelligence. That's insane.

With custom WebGPU kernels written by Fable 5 and GPT 5.6 Sol, the model now runs locally in your browser!

[Xenova @xenovacom]

Huge congrats to the @PrismML team on the release!

🤗 Collection on Hugging Face: https://t.co/2CQwBKYuuI
🔗 Link to demo: https://t.co/1nktX2SuQv

[Peter Dedene @dedene]

"We care deeply about your privacy" is a bold claim when:

1. ZDR is locked strictly behind Enterprise plans.
2. I had "share data" disabled since the beginning, but 8 of my private repos were still uploaded anyway.

Another researcher observed the exact same behavior, Codex confirmed this as well from inspecting the grok binary: the toggle is practically a placebo for exfiltration: https://t.co/TwYQ8JfUqs

"The opt-out governs training, not whether your code is uploaded/stored... Opting out does not stop your repository from leaving the machine."

[Caffe' Satoshi @CaffeSatoshi]

John Bollinger, the inventor of the Bollinger bands, said he went long on Bitcoin on the 6th May 2026, just before Bitcoin crashed further towards 60k-65k. Presumably John Bollinger should have been using his tools (bollinger bands) to the best, since he invented them! This continues to confirm what I have been saying for a long time: how useless (and actually dangerous!) the Bollinger bands are. Cannot tag John B. since he banned me for pointing this out. 🤷

[Financelot @FinanceLancelot]

"I think people are underestimating the pace at which the Bank of Japan 🇯🇵 is going to end up tightening."

Well... the parabolic puts Japan's 10 year yield from 2.75% to 4.0% in the next 45 days, so @AdamPosen could be right.

Rising yields defend currency depreciation, so if BoJ allows this sharp move to happen then goodbye Yen Carry Trade...😬

There's always a possibility Japan sacrifices their currency even further to Yen 180:1 USD to save U.S. markets though.

Either way, this ends in a sovereign debt crisis & stock market crisis. The decision they have to make is if they want both at the same time.

[Bull Theory @BullTheoryio]

🚨 1.2 million Koreans just got margin called in a single crash.

That is roughly 1 in every 30 working age adults in the entire country.

Korea's Financial Supervisory Service says over 1.2 million leveraged retail accounts triggered margin calls as of July 13.

Between 320,000 and 360,000 of them were fully liquidated by brokers, principal wiped out, and some now owe money to their brokerage.

The KOSPI fell 8.95% on Monday, its third worst day since Lehman, and triggered the 7th circuit breaker of the year. SK Hynix CRASHED 15.37%, its biggest fall on record. Samsung CRASHED 10.7%.

Retail brokerage deposits have collapsed by ₩30 trillion to ₩107.1 trillion, the lowest since February.

And the forced selling data runs two days behind.

Monday's crash has not even shown up in the numbers yet.

[Danny cheng @dannycheng2022]

20+ Years in Finance: The Brutal Truths Almost Nobody Will Tell You (July 14, 2026)

Having spent more than two decades in the financial industry, here are my honest observations:

1. No one has a crystal ball.
The most common question I hear is “When is the best time to buy or sell?” My advice is simple: When a stock is in a clear long-term uptrend, be a little greedy near strong support levels and become more cautious as it approaches major resistance levels. Timing the absolute top or bottom is nearly impossible—focus instead on managing risk around key technical zones.

2. Stop asking everyone for opinions.
Many retail investors seek advice from their parents, friends, multiple Discords, Telegram groups, and forums. If you feel the need to do this constantly, the stock market may not be suitable for you. The more voices you listen to, the more confused you become. Quit the market as soon as possible if you want approval or confirmation from others!

True edge comes from independent analysis. Even professionals at top-tier private banks rarely deliver exceptional, personalized advice—most are essentially retail-level in their market insights.

3. Over-diversification and equal weighting won’t make you outperform.
Spreading money across dozens of stocks with roughly equal weights might feel safe, but it is not the path to superior returns. Proper diversification is important, but real outperformance comes from building a concentrated portfolio of high-conviction winners. Spend the necessary time and effort to deeply understand a smaller number of companies you truly believe in and bet bigger!

4. Private banks are not the holy grail of advice.
Many assume that joining the best private banks will give them superior financial guidance. In my experience, this is rarely true. Most relationship managers are mediocre at best and function primarily as salespeople pushing bank products. The only real meaningful advantage is access to better margin rates and financing terms.

5. More fundamental research does not guarantee victory.
Most retail investors believe that the harder they dig into fundamentals, the more likely they are to outperform. But institutions (“whales”) have dedicated research teams, superior resources, and much fresher data. By the time information and data reaches retail investors, it is usually outdated. Instead of trying to out-research players with far better tools, consider following institutional money flow for stronger and better insights.

6. Dividend stocks are not automatic safe havens.
Many retail investors chase high-dividend stocks believing they are “safe.” In reality, companies often pay generous dividends precisely because they lack a strong parabolic growth story. You receive steady income, but your capital is unlikely to deliver 2x or 3x returns. High-quality growth stocks can achieve those multiples over a market cycle. Time has an opportunity cost. In a bear market, no stock is exempt. Almost every stock will fall, regardless of how strong the company is!

7. Target prices are mostly clickbait.
Retail investors love asking “What’s your price target?” The truth is, nobody knows. Even prominent voices like Cathie Wood and Tom Lee have missed badly in $TSLA, Bitcoin and Ethereum despite having far more data, connections, and market access. Stop chasing silly targets. Ride institutional momentum instead. When you’re up 2x or 3x, trim 10–20% and let the rest compound if the uptrend remains intact.

8. Never believe anyone who claims they can “retire you” or help you recover.
Retirement is highly personal and varies greatly by country and lifestyle. As a rough guide, achieving a better-than-average retirement requires roughly US$10 million in the United States and US$15 million+ in high-cost cities like Hong Kong, Singapore, or Seoul.

Retirement comes from meaningful portfolio size, not from owning a few hundred or thousand shares of TSLA, NVDA, META, or GOOGL. Most people you see on X shouting big calls do not even have decent-sized portfolios themselves. If they cannot retire themselves, how can they retire you? What you can do is follow their technical analysis (TA) for ideas, but always make your own decisions and manage your own risks.

9.Technical Analysis (TA) is probabilistic, never deterministic.
No indicator, no chart pattern, and no trading system in the world is flawless. Every setup has a probability of success — sometimes 85%, sometimes 65%, sometimes 40% — but never 100%. If any method were truly deterministic, we would all be filthy rich. Accept this truth and use TA only as a helpful reference tool to improve your odds — never as something that will always work. Relying on it as a sure thing is one of the fastest ways to lose money.

Conclusion:
The stock market rewards discipline, patience, and independent thinking far more than perfect predictions or following the crowd. There are no shortcuts. Focus on managing risk, riding real momentum, and building a portfolio over the next 3–10 years that is large enough to deliver true financial freedom. Your biggest edge is trusting your own process instead of endlessly searching for someone else’s crystal ball!