This Week in AI

Community & Newsletter for AI entrepreneurs, and operators!

[This Week in AI @thisweekinai_]

https://t.co/U7FS6ibSlD

[This Week in AI @thisweekinai_]

Fable 5 launched with a variety of safety mechanisms,

1) Training the model to decline to assist with dangerous requests; others involve retroactively analyzing patterns of misuse.
2) Another involves classifiers—smaller automated AI systems that, during an interaction, detect when the model is asked to perform a potentially harmful cybersecurity task (or produces potentially harmful outputs). When this occurs, the classifiers block the model from responding to requests. The ultimate goal of these classifiers is to prevent the model from engaging in uniquely dangerous behaviors.

https://t.co/Vv6lrfqIOv

[This Week in AI @thisweekinai_]

https://t.co/g4kgz1J3EX

[This Week in AI @thisweekinai_]

@AnthropicAI To win back approval, they made massive regulatory concessions, agreeing to co-develop future safety standards and report malicious activity directly to the feds.
https://t.co/UT7zSrTFVs

[This Week in AI @thisweekinai_]

https://t.co/g4kgz1J3EX

[This Week in AI @thisweekinai_]

@CuriosityonX How long until communities stop treating data centers like a golden goose and start treating them like a giant, thirsty roommate who needs to be evicted?
https://t.co/c3B76O3wji

[This Week in AI @thisweekinai_]

https://t.co/GhakAWXsN3

[This Week in AI @thisweekinai_]

Is there any real way out of this silicon squeeze beyond just waiting years for new mega-fabs to go live? Because if we can't solve this on the software side through extreme architectural efficiency, hardware inflation is going to completely choke out every builder who isn't a tech giant.
https://t.co/H7IxlQnaDd

[This Week in AI @thisweekinai_]

https://t.co/GhakAWXsN3

[This Week in AI @thisweekinai_]

@saeverley @axios It’s wild that all US data centres combined use less than a third of the water we throw at keeping golf courses green.

https://t.co/c3B76O3wji

[This Week in AI @thisweekinai_]

https://t.co/GhakAWXsN3

[This Week in AI @thisweekinai_]

@CuriosityonX It's wild that we're willing to burn through a very real, finite planet just to build a limitless digital mind.
https://t.co/c3B76O3wji

[This Week in AI @thisweekinai_]

https://t.co/GhakAWXsN3

[This Week in AI @thisweekinai_]

@PalantirTech We agree: https://t.co/w9ApQMfhnc

[This Week in AI @thisweekinai_]

https://t.co/Kx0UV0LdKo

[This Week in AI @thisweekinai_]

@sab8a Hey this is gold, thanks!

[This Week in AI @thisweekinai_]

https://t.co/GhakAWXsN3

[This Week in AI @thisweekinai_]

@natolambert

[This Week in AI @thisweekinai_]

Why?
Under a compute hour model, if Anthropic makes Claude 2x faster, Amazon gets the benefit because they need fewer server hours to process the same workload.

Under a token pricing model, Anthropic pockets 100% of those optimisation gains. As inference gets cheaper to run internally, Anthropic's profit margins skyrocket, while Amazon continues to pay a flat rate per token.

Now it's making sense!

[This Week in AI @thisweekinai_]

@Polymarket That's why 9 out of 10 of us have cancer!

[This Week in AI @thisweekinai_]

@mattshumer_ The better question is, can they stop cross-border inference API calls to Chinese LLMs?

[This Week in AI @thisweekinai_]

@WhaleInsider If China cloned the "restricted" capabilities of Claude Mythos anyway, did US export controls achieve anything besides giving our global rivals a massive head start?

[This Week in AI @thisweekinai_]

@idai_institute To avoid concentration and avoid dependency, This is the only way out !
https://t.co/b2nIS82P3v

[This Week in AI @thisweekinai_]

https://t.co/Kx0UV0LdKo

[This Week in AI @thisweekinai_]

@GuyZys Open source AI must win!
https://t.co/b2nIS82P3v

[This Week in AI @thisweekinai_]

https://t.co/Kx0UV0LdKo

[This Week in AI @thisweekinai_]

@BVeiseh The theme AI this week: Dependency !
https://t.co/b2nIS82P3v

[This Week in AI @thisweekinai_]

https://t.co/Kx0UV0LdKo

[This Week in AI @thisweekinai_]

@Coinvo But only for US organisations that operate and protect critical infrastructure, such as OpenAI 5.6, with limited access.
https://t.co/b2nIS82P3v

[This Week in AI @thisweekinai_]

https://t.co/Kx0UV0LdKo

[This Week in AI @thisweekinai_]

@bindureddy That's why every nation should have its own AI !
https://t.co/b2nIS82P3v

[This Week in AI @thisweekinai_]

https://t.co/Kx0UV0LdKo