David J.

Tech & AI innovation should cost less. Building products to make that happen. Opinions on AI, crypto, and European market developments.

[David J. @lordofblocks]

https://t.co/vQyFGVEF5v

[David J. @lordofblocks]

@nntaleb The funny thing is humans have the same problem.

Most researchers never see the full set of failed experiments either.

[David J. @lordofblocks]

@unusual_whales This is the kind of thing people mean when they say AI is changing companies.

[David J. @lordofblocks]

@BerntBornich Getting an AI to write code is one thing.

Getting a robot to find the milk and not put it in the bin is still the boss level.

[David J. @lordofblocks]

@DeItaone To get anywhere near those numbers, you'd need demand for inference to absolutely explode and stay strong even as costs keep falling.

[David J. @lordofblocks]

@emollick Some of it is also expectations moving.

The moment a capability becomes normal, people stop counting it as progress.

[David J. @lordofblocks]

@MatthewBerman Everyone supports a pause right up until there's a chance someone else keeps training in secret.

[David J. @lordofblocks]

@mark_k @OpenAI Persistent memory is one of the strongest moats in AI.

The more context an assistant builds up, the harder it is to switch away.

[David J. @lordofblocks]

@AnthropicAI If frontier models are improving the process of building frontier models, release cycles could compress a lot faster than people expect.

[David J. @lordofblocks]

@testingcatalog We're getting close to the point where ChatGPT remembers your habits better than you do.

[David J. @lordofblocks]

@curious_founder If compute can be brought online months faster, that becomes a serious competitive edge. Model progress is increasingly tied to infrastructure execution.

[David J. @lordofblocks]

@OpenAI We're getting closer to the point where your AI remembers things about you that you've completely forgotten yourself.

[David J. @lordofblocks]

@AndrewCurran_ Everyone agrees coordination would be nice.

The problem is nobody wants to be the one who pauses while the other guy keeps training.

[David J. @lordofblocks]

@mark_k Maybe, but we're still at the stage where AI is helping humans build better AI.

That's a long way from fully autonomous improvement.

[David J. @lordofblocks]

@scaling01 More code shipped doesn't automatically mean more value shipped.

Anyone who's worked on software knows those aren't the same thing.

[David J. @lordofblocks]

@Polymarket Hard message to sell when the same labs are racing to ship bigger and better models every few months.

[David J. @lordofblocks]

@PeterDiamandis The best kept secret of the last two years is almost all the gains have been software, not hardware. Everyone is still just counting H100s.

[David J. @lordofblocks]

@Dr_Singularity Feels a bit early to call it an intelligence explosion. What Anthropic showed is real, but it's still mostly AI helping humans build better AI.

[David J. @lordofblocks]

@_sholtodouglas A lot of dev work now feels like babysitting a team that never sleeps and occasionally gets very creative.

[David J. @lordofblocks]

@beffjezos Hard to blame anyone for keeping a tool in-house when that tool is helping build the next tool.

[David J. @lordofblocks]

@IterIntellectus The timeline debate gets a lot harder when the labs start showing internal data instead of just talking about the possibility.