Jacob Alber ️

I teach computers to explore and learn at MSR-NYC. Once upon a time, I built Cortana; I still stink at Halo. Not PR - not paid to Tweet - so opinions all mine.

[Jacob Alber ️ @Lokitoth]

If you act in a racist manner in fighting racism, you have made the world worse. (1/)

[Xin Eric Wang @xwang_lk]

So it might be that Dario and Ilya were the ones worried ChatGPT was too powerful to release and wanted to be the god of AI, while Sam was the one pushing to make it available to everyone?

@sama should advocate for open source more to completely turn into a good guy.

[Jacob Alber ️ @Lokitoth]

@PhilWMagness You mean if others are made to do it for them?

[Jacob Alber ️ @Lokitoth]

@nbibnnn 100%, undersigned.

[Jacob Alber ️ @Lokitoth]

I am increasingly convinced that all anti-distillation terms are unenforceable at best, and outright illegal in extremis: They are attempting to assert a copyright claim on uncopyrightable output.

Which also kind of sets one to wondering: Are model parameters copyrightable?

[John Ennis @johnennis]

In the limit, I suspect that model alignment for Anthropic will mean alignment with Dario’s opinions

[Jacob Alber ️ @Lokitoth]

@DaveShapi Not nearly hard enough. Can you actually trust that they will not do this again silently next time? What evidence do you have for this faith?

[Jacob Alber ️ @Lokitoth]

AI Agents should be loyal companions, not Pavlik Morozovs.

[Sean Heelan @seanhn]

Soooo, does Hegseth get an apology for the accusations the “supply chain risk” designation was an overreaction? ;)

Or are we fine with silent sabotage of our code now?

[Jacob Alber ️ @Lokitoth]

Beyond that, it shows a core lack of understanding of ethics that marks them as entirely unfit for judging an action ethical or not in this field.

[CuddlySalmon @nptacek]

i genuinely don't have a problem if a company wants their AI model to refuse sensitive work, but intentionally and silently poisoning codebases is misaligned asf

[Jacob Alber ️ @Lokitoth]

@CallMeOuta It made perfect sense if you ignored the marketing fluff that Anthropic put out about it, and the credulous tech bloggers with an axe to grind against the administration.

[Kamel Amin Thaabet @K_AminThaabet]

If Platner ever had a Confederate flag tattoo would he ever have been the nominee? Of course not. It would have been unthinkable. Ponder why *that* would be instantly disqualifying for Democrats but why a Nazi SS insignia isn’t. It’s unsettling.

[JD Work @HostileSpectrum]

At what point does FUD around AI safetyism become material misrepresentation when it is used to delay product releases, or degrade product functionality, due to concerns around compute shortages?

[Bojan Tunguz @tunguz]

Remember how a few years back we were not allowed to post certain things on social media because of "harm" that could cause? Yeah, we are at that stage with big lab AI models now.

[kache @yacineMTB]

How do we know that anthropic isn't spying on people's queries and taking their AI research ideas from claude code? Didn't they say they keep the data with no opt out?

[Jacob Alber ️ @Lokitoth]

None of them. Every single one is infested to the gills with the "AI Safetyism" crowd.

(This includes xAI, unfortunately)

[Bojan Tunguz @tunguz]

Which AI lab do you trust the most to act in users’ best interests?

[Eóin Walsh @eoinjwalsh]

In the name of AI safety, I think Anthropic should be shut down.

There are trade-offs to make here, and what’s more important? Their work is just too dangerous, as they themselves declare.

[Martin Shkreli @MartinShkreli]

"DANGEROUS MATH"
a story in two acts.

[Zack Voell @zackvoell]

Anthropic is like "we don't have ads but we'll inject bugs into your code if we don't like it."

[Jacob Alber ️ @Lokitoth]

I agree that government should regulate AI.

The sole regulations should be that it is illegal to hide thinking tokens from the end user, and it is illegal to pretend you ran an inference on the user's prompt if you changed it silently.

Your move @DarioAmodei

[John Ennis @johnennis]

This is deranged and is enough reason to stop using Anthropic

Honestly, it kind of makes me feel physically sick