Most people have no idea what is happening. (AI is crazy good at some things.)
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Prin retweeted
Mám za sebou prvních cca 300 km testování Tesla samořízení v ČR. Centrum Prahy, dálnice, okolí.
Mých prvních TOP 20 dojmů:
1. Je to wow. Mnohem lepší, než sem čekal. A to mám nakoukaný videa z Ameriky. Za 5 minut sedíte jak v taxíku. Ta rychlost "zvyknutí" je neuvěřitelná.
2. FSD umí jezdit rychle. A někdy i nad rychlostním limitem. Pořád je to bezpečná jízda, žádný manévry na Turka. Ale prostě to umí přetáhnout.
3. Překročení rychlosti se dá ručně štelovat nebo úplně vypnout. Pak se drží 100% správné rychlosti. Ale zase vás vytroubí každý Bolt v centru na třicítce.
4. Na radar sama nezpomalí. Přestože to má v mapách. Prej umělá inteligence, ale pokutu dostanu já, že.
5. Neumí odbočit doprava na zelenou doplňkovou šipku.
6. Uzavírky v centru jsou problém. Žluté čáry, blbé značení, změny směrů... je z toho evidentně nervózní a já s ní. Někdy je to značení naprosto kokotský. V jednom úseku, kde bojovala, jel za námi lidský řidič a bojoval úplně stejně, možná hůř.
7. Dálnice naopak fungují na 100 %. Pouští auta v připojováku, a to s předstihem. Okamžitě reaguje na kamiony, když se začnou předjíždět. Řadí se sama do pravého. Umí si zrychlit, zpomalit podle potřeby. Dálnice je pocitově vyřešená.
8. Neumí BUS pruhy spolehlivě. Chová se, jako by tušila, že je něco špatně, ale nic s tím neudělá. Jako děcko, co ví, že něco provedlo, ale bojí se to říct.
9. Reakce na chodce neuvěřitelně dobrá. Někdy až přehnaná. Boreček stojí na červené na přechodu, zvedne oči od mobilu a přešlápne – Tesla okamžitě jemně přibrzdí "kdyby náhodou". To je v rámci bezpečnosti super.
10. Zvládá krásně tramvaje, dává přednost, hlídá si je. Ostrůvky tramvajové ok. Jednou si ho všimla na poslední chvíli a trošku s tím cukla. Ale nic zásadního.
11. Při změně pruhů v malé rychlosti někdy "plave" zleva doprava a cuká rychle volantem. Extrémně nepříjemné. Vypadá to jako bug. Možná je to jen u mě. Dejte prosím kdyžtak vědět.
12. Umí si krásně objet překážku. Delivery dodávka blokuje cestu, počká si, blinkr, objede. Bez problémů.
13. Odbočovací pruhy jsou trošku divný v centru. A to sem jel večer a nebyla doprava. Zítra zkusím ve špičce. Ale nedivím se. Na některých křižovatkách v Praze musíte řazení plánovat s předstihem. Mnohem dříve, než jde vidět značení. Kdo tam jede poprvé, diví se a dělá zmatky na poslední chvíli. Tesla tam vždy jede jako poprvé.
14. Ale Anděl dala bez ztráty hvězdičky. Skoro. Křižovatku Ostrovského–Vltavská projela pochopitelně jako: levý blinkr – pravý blinkr. No, jak Brňák v Praze na návštěvě.
15. Reakce na retardéry trošku retardovaná. Chtělo by to více zpomalit. A najíždět jemněji. Mám rád svoje tlumiče.
16. Zvládá naprosto s přehledem úzké uličky a malé prostory. Jakože fakt jede sebevědomě na centimetry jak starej autobusák.
17. Na parkovišti před obchoďákem umí přijet k bráně, počkat na lístek a jet, až se brána otevře. Jen lístek musíte vzít vy.
18. Vnitřní kamera vás buzeruje, a to dost. Stačí vzít do ruky mobil na červené a okamžitě na vás řve. Musíte koukat před sebe. Ne ven, ne na displej, ne na spolujezdce. Před sebe.
19. Nemusíte mít ruce na volantu. Můžete mít volně položené v klíně, na opěradle nebo mít překřížené. Pokud se ale začnete protahovat, dávat ruce nad hlavu, hned jí to vadí.
20. Verze v EU má některé vychytávky navíc. Třeba povinnost oznamování o změně pruhu s předstihem. Což je naprosto idiotský, protože místo na cestu čumíte na text na displeji. Bere to pozornost. A ještě je to nespolehlivé, protože překvápko: doprava je dynamická a to, že auto chtělo změnit pruh před 5 sekundama, neznamená, že ho změní teď. Tak jen doufám, že to úředníci nebudou kurvit dalšíma podobnýma vylepšeníma.
Jinak je to super. Sem z toho nadšenej. Sci-fi technologie totálně. Budu testovat dál. V pátek si řekneme v Schengenu v TECHGUYS další detaily.
Czechia joins several other EU countries that approved Tesla FSD Supervised ahead of EU-wide vote.
🇨🇿 Asistenční systém Tesla FSD Supervised bude možné za stanovených podmínek používat také v Česku. Uznali jsme prozatímní schválení nizozemského úřadu RDW. Připojujeme se tak k Nizozemsku a dalším pěti evropským státům.
Tomuto kroku předcházely měsíce odborného posuzování a jednání s českými i zahraničními odborníky a zástupci Tesly. Zajímalo nás především, jak systém funguje v běžném provozu, jak reaguje na dopravní značení a rychlostní limity a jak zajišťuje, aby řidič řízení stále věnoval pozornost. Důležité byly také zkušenosti ze zemí, kde už systém používají.
Chceme moderním technologiím otevírat cestu na české silnice. Bezpečnost přitom zůstává na prvním místě.
ℹ️ Co to znamená pro řidiče? Tesla FSD Supervised je stále asistenční systém. Řidič musí sledovat provoz, dohlížet na jeho činnost a být připraven kdykoliv okamžitě převzít řízení. Odpovědnost za jízdu zůstává plně na řidiči.
👉md.gov.cz/Media/Media-a-tisk…
Prin retweeted
It’s not news to me that Shai and his team move at lightning speed. But man… how old is @typesafeai Jev? 4 days? Holy moly. They don’t waste any time.
For those who do not know what this is:
Jev = ultra-fast AI that doesn’t write text. It answers structured questions (classify / score / true-false) so software can decide instantly. $LMND
Lemonade OS now runs @typesafeai Jev. For the right use case, it’s really, really fast.
Prin retweeted
The last week has really shown me that someone who wants to understand AI risk has no good place to start.
Hence we made agi.fyi, the Wirecutter for content about AI Risk. We're launching with 3 articles: 🧵
Prin retweeted
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
Prin retweeted
It’s absolutely ridiculous that some people are suggesting that out of these two, the one who should be distrusted because his claims are in his supposed self-interest is Coxon
Prin retweeted
Options + leverage have turned the stock market into a casino.
The huge rally off the March low in semis made little sense and now the degrossing + panic selling is also getting ridiculous!
The fundamentals have remained relatively steady, just sentiment has done a U-Turn.
Prin retweeted
This is a super exciting release - Claude Fable 5 is the same underlying model as Mythos but with added safeguards. The benchmarks are great and it's SOTA on everything by a margin but I'll add that *qualitatively* also, this is a major-version-bump-deserving step change forward (imo of the same order as Claude 4.5 was in November), peaking especially for long problem-solving sessions on very difficult problems. You can give it a lot more ambitious tasks than what you're used to, the model "gets it" and it will just go, and it's never felt this tempting to stop looking at the code at all (but don't do this in prod!). The model still has quirks that people will run into and the safeguards are configured to be a little too trigger happy for launch, which can hopefully be tuned over time.
I feel a lot of things changing as working software increasingly comes out on a tap. The Jevon's paradox kicks in and I feel my own demand for software growing substantially. You can ask for anything - explainers, visualizers, dashboards, bespoke single-use apps (e.g. a full wandb that is hyper-specific just for your project), you can 10X your test suite, auto-optimize code, run giant research projects with custom HTML for the results, anything! "Free your mind" (Matrix ref). Really looking forward to all the things people build!
Replying to @claudeai
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision.
The longer and more complex the task, the larger Fable 5’s lead over our other models.
Prin retweeted
In this new Berkeley eval of AI for zero-days @Aisle_Inc is #1 globally in 3 out of 8 categories:
1) total zero-days with CVEs (!!)
2) breadth of vuln types
3) coverage of most dangerous vuln types
We're beating both Google and Anthropic with a few dozen people out of Europe!
Agents are finding more vulnerabilities than ever. But it turns out there are gaps in existing vulnerability discovery. Over the past 90 days vs. a year ago, web vulnerabilities (XSS/SQLi/CSRF) are down 66% and memory safety exploitability is down 3.5x.
We built the Agentic Vulnerability Coverage Map to track it all, updated daily. Introducing the Berkeley Vulnerability Initiative: vuln.cs.berkeley.edu. ⤵️
7.5% of cars sold in April in 🇨🇿 were EVs 🔋. Not bad.
Czech Republic reported 175 Tesla sales and 0.8% market share in April. BEV penetration is 7.5% and Tesla has 10.4% of this segment. 🇨🇿
• Market share is 0 basis points or 0% above the 3-month trailing average of 0.8%
• 93% Model Y and 6% Model 3
• +3% vs. April last year and +77% compared to January the first month of the previous quarter
• Second best April ever
• 4th best first month of the quarter ever and +77% vs. the previous one
• Best month since 11 months
• Highest first month of the quarter since 24Q4 (6 quarters)
• Last three months +50.8% vs. November - January
• Year-to-date -5% over same period last year
• Year-to-date is 34% or 4.1/12 of last year's total
The following article was trending on Bloomberg yesterday, and it was likely one of the biggest drivers behind the software selloff (which also took down LMND, even if it doesn't make much sense).
citriniresearch.com/p/2028gi…
Prin retweeted
12 Nobel Prize winners have warned that we may someday lose human control over advanced AI systems
- Acemoglu (Economics '24)
- Hinton (Physics '24)
- Hassabis (Chemistry '24)
- Bernanke (Economics '22)
- Parisi (Physics '21)
- Ressa (Peace '21)
- Doudna (Chemistry '20)
- Bengio (Turing Award '18*)
- Fihn (Peace '17 via ICAN)
- Mather (Physics '06)
- Wilczek (Physics '04)
- Stiglitz (Economics '01)
*equivalent to CS Nobel
Prin retweeted
Slováci mají mnohem efektivnější spoření na důchod než Češi. Průměrný klient v tuzemsku musí vynakládat až 10 x víc než Slovák, aby měl stejnou rentu. Nová studie @FPertold a @skejwin z IDEA při CERGE-EI srovnala výnosy v ČR, na Slovensku a ve Švédsku.
seznamzpravy.cz/clanek/ekono…
Prin retweeted
🚨🇩🇪 GERMANY JUST WENT LIVE WITH TESLA FSD
Germany has launched the first government-backed FSD (Supervised) shuttle service in the EU, running in the Eifel district of Bitburg-Prüm with full support from regional transport authorities.
The pilot project was openly accompanied by the Rhineland-Palatinate Minister of Economic Affairs and Transport, plus local officials.
The result is a real gain for rural mobility: greater accessibility, more flexibility and tangible benefits for everyday life.
Europe is moving from endless autonomy debates to real-world deployment, signaling regulatory trust and momentum that could ripple across the EU.
Quietly, this puts supervised autonomy on the road with government approval, not resistance.
Source: @alex_avoigt, @Tesla
Prin retweeted
Latest podcast from @Gregory_C_Allen has an insane section on criminal activity at Meta.
Internal docs leaked to Reuters show:
• 10% of all Meta revenue comes from ads for scams & banned goods ($16B/year)
• Meta estimates it's involved in 1/3 of all successful scams in the US
• That suggests they drive $50B in scam losses for US consumers alone each year
• Meta earns ~$3B annually from scam/banned goods ads run by Chinese operations alone
The China case study:
• In 2024, Meta made $18B+ from Chinese companies advertising to foreign consumers
• Internal teams found ~19% was scams/banned content
• An anti-fraud team successfully cut these ads in half
• When Zuckerberg saw the revenue impact, he told them to "pause" and the team was disbanded
• By mid-2025, banned ads climbed back to 16% of China revenue
• This results in money being stolen and going directly from ordinary Americans to Chinese criminals
The deliberate enabling:
• Fraud earns 10% of all revenue, but anti-fraud teams were blocked from any action costing >0.15%, so they couldn't effectively do anything
• Meta charged higher rates for suspected fraudulent ads — a "scam tax"
• Their algorithm naturally identifies people vulnerable to frauds and feeds them more and more
The cold calculation:
• Meta anticipated up to $1B in regulatory fines for this
• But they make $3.5B every 6 months from high-risk ads
• They view these fines as just "cost of doing business"
Senators Blumenthal & Hawley now calling for FTC/SEC investigations in a blistering letter, noting that all this happened while Meta cut safety staff and moved billions over to VR and AI.
WTF.
Prin retweeted
So after all these hours talking about AI, in these last five minutes I am going to talk about:
Horses.
Engines, steam engines, were invented in 1700.
And what followed was 200 years of steady improvement, with engines getting 20% better a decade.
For the first 120 years of that steady improvement, horses didn't notice at all.
Then, between 1930 and 1950, 90% of the horses in the US disappeared.
Progress in engines was steady. Equivalence to horses was sudden.

































