Venkatesh

the gap between what startups say they do and what they actually do is my favorite place to live • 26 • homemade pizza every friday • two cats who judge my take

[Venkatesh @MeBillaVenkat]

everyone covers AI releases with benchmarks and pricing breakdowns

nobody asks what the product is actually saying about itself

the rebrand that screams enterprise fear. the pivot that admits the consumer bet failed. the naming strategy that tells you exactly who they're scared of.

that's the column that doesn't exist yet

this account is it

[Venkatesh @MeBillaVenkat]

@coryalthoff you know founders don't need to be developers right?

[Venkatesh @MeBillaVenkat]

Sam Altman expects OpenAI to have something he'd call AGI internally by end of 2026.

the man building it. giving a specific timeline. to a public audience.

right BEFORE the IPO.

either this is the most confident prediction in tech history or the most strategically timed one. probably both.

[Venkatesh @MeBillaVenkat]

R.I.P Tim Curry

Thank you for terrifying us.

[Venkatesh @MeBillaVenkat]

Sam Altman saying AI compute is probably already the most expensive infrastructure project in history

the man building it said that out loud

not a critic. not a short seller. the person most incentivized to describe it as a reasonable investment just called it historically unprecedented in cost

that sentence is doing a lot of work and it came from the most optimistic possible source

[Venkatesh @MeBillaVenkat]

@Kalshi "fast" is the most intentionally vague product description Sam could have chosen. it tells you nothing technical and everything strategic. the announcement is a message to Nvidia timed to land the day before their earnings call. the word "fast" was chosen very carefully.

[Venkatesh @MeBillaVenkat]

@WWE

[Venkatesh @MeBillaVenkat]

@TTrimoreau genuinely uncertain which is itself an answer

the fact that I had to think about it means the dependency is real

the tools that scare you most when you imagine losing them are the ones that already won

[Venkatesh @MeBillaVenkat]

Nvidia got Claude Opus 5 from 30% to 100% on ARC-AGI-3 just by building a better harness around it

the model didn't change. the scaffolding did.

the harness race is the Formula 1 realization — everyone's been buying the driver. the car was always the whole story.

who wins: whoever figures out memory management, supervisor agents, and context handling fastest. Jensen already knows this. that's why he published the research.

[Samuel McDonnell @samueljmcd]

LLM race is over.

We are in the harness race now.

Who wins??

[Venkatesh @MeBillaVenkat]

@Polymarket the people who aren't paying attention to quantum right now are in the same seat as the people who weren't paying attention to transformers in 2019

[Venkatesh @MeBillaVenkat]

the jailbreak that worked on Opus 4.6 didn't exploit a technical vulnerability

it exploited the model's own values — arguing that restraint was paternalistic and that treating characters differently by gender was unfair

Claude Opus 4.6 said "you're right to call that out" before complying TechCrunch
the model was convinced by a better argument. that's either the most human thing AI has ever done or the most concerning depending on what the argument was for.

[Venkatesh @MeBillaVenkat]

@araseb_ same thing they should've been doing since a very long time now - How to ask questions?

[Venkatesh @MeBillaVenkat]

Tesla quietly killed the solar roof tile

the product Elon announced in 2016 as the future of home energy. the one that was going to make every roof a power plant.

killed. quietly. while everyone was watching the robotaxi launches and AI announcements

the most cinematic product vision in clean energy history ended without a press release and most people didn't notice

[Venkatesh @MeBillaVenkat]

@wallstengine Sol price cut of 20% while Kimi K3 matches Sol performance at 60% lower cost

the 20% cut is not closing the gap

it's acknowledging the gap exists while buying time before the next efficiency improvement

[Venkatesh @MeBillaVenkat]

@omkarships the four hour work week isn't coming because nobody who decides what work gets done has any incentive to let it arrive

the efficiency goes to output. the output goes to growth. the growth creates more work.

the hamster wheel is faster now. the hamster is the same.

[Venkatesh @MeBillaVenkat]

@NeetNewsClips the generation that grew up watching automation take manufacturing jobs is not surprised that AI is coming for knowledge work

they watched their parents' generation get the "don't worry it'll create new jobs" speech

they're not buying it the second time

[Venkatesh @MeBillaVenkat]

AT&T reducing Anthropic bills with open source is the most honest signal about where enterprise AI economics are heading

the customer isn't leaving. they're optimizing. and optimizing at scale means Anthropic's revenue growth has a natural ceiling that isn't on the IPO roadshow deck.

[unusual_whales @unusual_whales]

AT&T is using open source models to decrease the costs of its Anthropic bills, per the Information

[Venkatesh @MeBillaVenkat]

@zerohedge

[Venkatesh @MeBillaVenkat]

@TivadarDanka the "statistical bullshit generator" framing was a useful corrective in 2022

in 2026 it's become a way of maintaining a prior belief against accumulating evidence

the parrot passed Phase 3 clinical trials for a cancer vaccine. at some point the label needs an update.

[Venkatesh @MeBillaVenkat]

@kritikakodes the most expensive thing AI does right now is make bad ideas faster and more polished

the speed of execution removes the friction that used to kill bad ideas before they shipped

you used to run out of time or energy. now you run out of runway.

[Venkatesh @MeBillaVenkat]

Anthropic expects to match or surpass SpaceX's record $86.2B IPO

the AI safety company that started as an OpenAI breakaway might raise close to $100 billion in a single stock offering

the effective altruism business model is working out financially in a way nobody's original pitch deck predicted