How many times does the world need reminding that leetcode bros (whether human or AI) do not make good engineers - literal inverse proportionality. Humans+AI collaboration is the way forward for the foreseeable future.

@ralexEmperorNeo @amoufarek Much better and way cheaper

@osanseviero Do you also plan on starting a DeepMind specific partner program to help deploy solutions at enterprises?

@jasonleowsg Wait wtf is a "senior" intern? 💀

@Ubunta @duckdb How do you manage deletes? How do you manage access control? For HA,It would be convenient to use duckdb as a query engine and not a DB.

Reality check:- - Deep research is mid. - o3-mini is mid too.

@DaveShapi Soo.. AI can build Ethereum by the end of 2025? 😂😂

@rauchg @vercel Bangalore?

@MaximeRivest Is this David Attenborough? 😁

@shl Did we have more infra/devops engineers before the cloud or after? That should give you a good idea

@prajdabre1 *naav

@NathanWilbanks_ @matsonj Aren't you still doing the drag and drop? Isn't there still a finite set of integrations that are pre-made?

@NathanWilbanks_ @matsonj This is like 3 python class files having 3 functions each which call each other on some condition. It's still using LLMs to do some series of intelligent acts, but I don't see what's agentic about it

@xlr8harder @ylecun Reasoning models != Auto regressive-Generative

@matsonj Yes.. Agents are being marketed from the perspective of absolute noobs. They're just chaining functions and slapping on a workflow UI on top of it. Gives the illusion of new capability

1000s of red teaming hours later:- 1. Refusals have gone up drastically for my marketing use-case 2. The most expensive (non-reasoning) API in the market, now has an additional 24% compute overhead 3. Still broken by elder plinius < 1 day Effective de/acc going on over there

RT @AnthropicAI: New Anthropic research: Constitutional Classifiers to defend against universal jailbreaks. We’re releasing a paper along…

@himanshustwts This is straight up BS. Absolutely smells of a scam