issaco

The best way to predict the future is to invent it. - Alan Kay.

[issaco @issaco]

@BusinessInsider Too cheap.

[issaco @issaco]

@firstadopter Not as bad as in Australia, how much tax is in this ? It should be $140-$170k

[Dan Luu @danluu]

There's no reason for software to be slow anymore:

https://t.co/H59Khcg0Ae

[issaco @issaco]

@ghhughes It’s so much worse when they do this in google workspace

[clem 🤗 @ClementDelangue]

So proud of our security team! They caught, contained & publicly disclosed an attack unlike anything we've seen before, and did it at record speed.

Also massively grateful to @Zai_org: they shared GLM5.2 as open weights (for free!) with the world and it became a key part of our defense.

This is day one for cybersecurity in the age of agents & we're all learning that secrecy is not the answer & that all defenders (not just a few selected ones) everywhere need more powerful models without restrictions, especially open ones!

[Perry E. Metzger @perrymetzger]

Defenders need to have access to good models so they can close security holes permanently. The incorrect model people work off of is that there is an unbounded number of security holes in the world, and that if you get smarter and smarter AI, that means you’ll just keep finding security holes forever. That’s not true.

If you give everyone developing software access to good AI models, they will close off the holes, and eventually, the well of security vulnerabilities dries up.

AI isn’t a crisis for computer security, it is a solution. The best hope we have of climbing out of the computer security crisis that we have been in since 1988 is letting developers fix the problems using AI systems.

[Tim Sweeney @TimSweeneyEpic]

One can imagine an alternate reality of the 1990's when 3D games rapidly emerged if the makers of the era were run by brazen opportunists who employed teams of lobbyists seeking regulatory capture: scare the world into thinking 3D graphics might be an existential threat.

[issaco @issaco]

@DavidSacks @stevesi Soon everyone can download advanced cyber defence and offence to test their defence with.

[issaco @issaco]

@mooreslawisdead It’s impossible to download your PS5 saved games for some regions, cancel and lose your cloud saves which are impossible to backup (unless something has changed) so cancel and lose your saved games ?

[issaco @issaco]

@ramez Unfortunately the only thing capable of running the deep pervasive constant surveillance required to police this was an advanced AI system…

[issaco @issaco]

@CockburnPol please confirm your outgoing number is not being used for spam calls

[issaco @issaco]

@lilyraynyc Google shopping is worse

[alphaXiv @askalphaxiv]

As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development

"Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning."

Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing.

This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider.

That is not safety. Safety policies should be transparent, auditable, and user-visible.

On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.

[issaco @issaco]

@gfodor Any prompt modification should be disclosed so we can QA and see what they are corrupting it into.

[issaco @issaco]

RTX spark #msbuild

[issaco @issaco]

@Smirkley Now do cotton

[Marc Andreessen 🇺🇸 @pmarca]

Co-sign.

[issaco @issaco]

@shiri_shh Microsoft’s only innovation will be to require CALs for each AI agent.

[issaco @issaco]

This new LM https://t.co/U9Oq9w7U3C is pretty fun, #bicycleface

[issaco @issaco]

The only good copilot … farewell

[GitHub Changelog @GHchangelog]

New signups for Copilot Pro, Pro+, and Student plans are paused to maintain service reliability for current users.

• Usage limits tightened; Pro+ offers 5X higher limits than Pro

https://t.co/sV8aOkGLZ9

[issaco @issaco]

The biggest AI risk isn’t that it gets too smart it’s that it gets too dumb to save costs (or because of model corruption, lack of ECC), decent powerful models silently downgraded or quantised without any oversight or knowledge of the actual integrity for the consumer of the API