tecstac

Fan of low tech

[tecstac @tecstac]

@unusual_whales You supported Trump and probably still are, so what's your point

[tecstac @tecstac]

@0xSero @foley2k2 There are 1000 low-quality books to every good book. When you're reading books, this is what you're reading. Just noise.

[tecstac @tecstac]

@lesliedeanbrown @WIRED omg, supermarkets are killing people! Why do you hate supermarkets so much? Where else would you buy groceries when you ban them to slow down jihadi mfs?

[tecstac @tecstac]

@kentcdodds How'd you protect your code bases against malicious code produced by agents? Code that can hide itself + spread further via prompt injection. Every reviewing AI may fall victim to that. Or do you have AIs that can't be tricked because they separate data+code? Which ones?

[tecstac @tecstac]

@Scott_Helme Technically correct, but some may remember the lockdown that happened with supported OpenID providers

[tecstac @tecstac]

@e_tews @hashbreaker Terminates at a geometric rate

[tecstac @tecstac]

@DevShah__ @bgurley As said, removing entirely would alter the data. You'd lose a lot of otherwise useful data

[tecstac @tecstac]

@KimZetter I just posted there about this.

Another reason could be to protect users from aggressive pushback / counterattacks from the models, but then they could've warned the users about that. But this might create more FUD.

We'd only find out from leaks.

https://t.co/v0ZLZiVQyW

[tecstac @tecstac]

@DevShah__ @bgurley They could filter everything out or rephrase everything, but that would alter the data with Claudish, for example, making it more biased than when the sources rephrase themselves

[tecstac @tecstac]

@DevShah__ @bgurley They could filter everything out or rephrase everything, but that would alter the data with Claudish, for example, making it more biased than when the sources rephrase themselves

[tecstac @tecstac]

Elon Musk is now officially an enemy of the President of the USA

https://t.co/Q8t1KmVdUh

[Elon Musk @elonmusk]

Excited to announce that all super intelligence organizations have now jointly agreed to the ultimate in AI/SI safety:

Moving all testing to Delta Airlines flights, where accessing the Internet is utterly impossible!

[tecstac @tecstac]

@MrFiberNet @0xSero Use of DRAM cache means the opposite. Once the cache is full, rates are going down because of cache misses. So unless your data fully fits in the RAM, this is unsustainable by definition.

[tecstac @tecstac]

In his previous life, immortal James Cameron was known as a surgeon

https://t.co/a86BSNUrU7

[tecstac @tecstac]

@VideoGameGun @0xSero Cerebras has articles explaining how it does binning

[tecstac @tecstac]

@TommySiegel @robertgraham Dune know about you, but commercially, it makes sense

[tecstac @tecstac]

@TommySiegel @robertgraham Wait until someone from Star Trek travels back in time, a long time ago, to a galaxy, far, far away.

[tecstac @tecstac]

@JeromySonne OpenAI's so gonna burn Anthropic's server racks totally accidentally while evaluating an internal model. Anthropic doing the same with OpenAI's server racks. Unsure whose racks Google's gonna target. Microsoft? Or rather Google DeepMind and Gemini?

[tecstac @tecstac]

@SwiftOnSecurity Interesting response text characteristics. Ask some more questions, and you can discern and fingerprint members

[tecstac @tecstac]

@SwiftOnSecurity Social engineering social engineers. How meta