Jimmy 🇺🇸🇺🇦

Security nerd. FSD nerd. Space nerd. Just generally a nerd.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

My AI research site, including (among many other papers) an examination of kimi's built-in bias:

https://t.co/M8eHf3xZC4

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@PhongGT @OmarShahine It was built on openclaw, with open collaboration between the two teams. So more of a coopetititor. Muse just straight up ripped off openclaw, though they did at least give some credit in an x post (it's open source, so nothing wrong with that).

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@Trajan230970 @OleTimeHardball And both were drafted as shortstops!

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@BBGreatMoments Just shared this with my dad who is in the ER with afib. He was born in 1939. He knew immediately who it was. His face lit up. Thank you for this, he needed it.❤️

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@SteamboatLion @petergyang I'm amazed at how few people understand this.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@ns123abc Serious question: can anyone point me to exactly where the AI companies actually asked for liability waivers? I can't find it. I've heard a lot of people saying they're asking for liability waivers. But i haven't heard them actually asking for it.

[Joseph 💎✌️🪑🇺🇸 Tesla Long Term Investor @ShrimpTeslaLong]

Dear @Tesla

👉 Toyota hates you because you are far more reliable.

👉 VW hates you because of your superior software infrastructure.

👉 Mercedes hates you because they can no longer claim the luxury title.

👉 BMW hates you because they are no longer the ultimate driving machine.

👉 Waymo hates you because you outperform them in safety, cost-per-mile, and aesthetics.

👉 Gas stations hate you because you don’t need gas.

👉 Dealerships hate you because they can’t charge a markup.

👉 Unions hate you because you take away their power and return it to the workers through meritocracy and incentives.

👉 Big government hates you because you show all of us how incompetent, outdated, and inefficient they are.

But with all that said, you do you, Tesla. Because you are changing the world.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

The problem with this has been terrible reporting and overreaction from people who are not actually reading what the the "alarmists" are saying. If you take the time to actually read Dario's paper, he is not asking for anything unreasonable, let alone irrational! He is saying let's take a breath and implement joint safety testing. That's it. It's perfectly reasonable, perfectly rational, something all the labs having been saying for a long time would have to happen. The narrative that this is irrational fear mongering or regulatory capture has got to stop. It's the reporting on this that's been irrational and irresponsible, not the ask itself. I wonder how many commenting about this have actually read Dario's paper?

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@Sinney_Scwasp @xLexemeX @elonmusk You said exactly what i was going to say!

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@edgecase411 I agree with every word of that post, which is a rare thing for me!

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@steipete It's about time! What took them so long? I wrote one for protonmail back in February. Or more accurately, my "Bucky" openclaw wrote it.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

You've got a lot of this really, really wrong. Which is unfortunate because there are parts that you also got right. But first of all, all of the AI companies offer Zero Data Retention options that keep your data off their cloud servers. There are also options available through both Azure and AWS that allow you to protect your data. Second, no one, absolutely no one, is suggesting that open weights models should be banned. Certainly not Dario or Sam. You really should do some research before posting trash like this, and maybe read what they're actually saying instead of just echoing other social media posters who also didn't read what Dario and Sam are actually saying. It's important because open weights models and owning the hardware will absolutely be the right answer for the largest law firms (like these folks) and some large healthcare entities. It's not anyone's worst nightmare, it's the right answer if you have (a) the money and (b) the expertise to manage this for yourselves. It's really hard for people who are trying to learn about AI to find accurate information and sensationalist posts like this doesn't help them at all.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

As a healthcare CISO, i can say that the fear outlined in this post is very real. It's not hype. This post by Kurtz is accurate without being doomerism or overly dramatic. The issue for the defenders is that even before the recent AI boom we were already able to find many more vulnerabilities than we were able to remediate. Now we're expected to defend 24/7/365 and have 100% of our infrastructure fully patched against even medium and low vulnerabilities (because AI can now chain those together making 3 mediums = 1 critical). And we're supposed to do that without introducing instability to the businesses we're trying to defend, maintaining 99.99+% uptime for our users? Agent swarms for defense are great in theory, but when the only way to really defend against what's coming is to be constantly, like at least daily if not hourly, applying all the latest patches? Many of us are lucky if we get a 4 hour window once a month to patch. If anyone figures out how to actually do this please let me know.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

Thank you @DarioAmodei this needed to be said. As a CISO for a healthcare company, I'm rooting for anything that might give us good guys a little more time. The fact that @elonmusk @sama and others are supporting this will hopefully give the skeptics and cynics pause. Let me know how I can help!

[Demis Hassabis @demishassabis]

Dario's essay points towards the right path forward. The details need working through, but the direction is correct for meeting this critical moment.

This is also why we recently put out our proposal for an industry-wide standards body for frontier AI. https://t.co/Mm1hmcaSmH

[Jimmy 🇺🇸🇺🇦 @because_sanity]

Alex, this is a rare case when i disagree with you. Hypothetically, if they were seeing things in the new models that they found genuinely terrifying, things that they felt might lead to the destruction of life on earth (or even just "irreversible damage"), what actions do you think they should take? Hypothetically, of course.

I still agree with you on most everything else. Mercury does indeed have it coming.

[Elon Musk @elonmusk]

Dario is right

[Jimmy 🇺🇸🇺🇦 @because_sanity]

That's hilarious. 250 people. I work in healthcare for a $250m company. I asked for two people in next year's budget (no chance at getting anything this year) to work on vulnerabilities. I got laughed at. I needed to "think outside the box" they said. This is reality for those of us who live in the real world, you know, the ones that have to make ebitda every quarter. Defender's window my ass.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

A bigger issue is going to be abliterated open source models, especially as quantized versions that run on small local setups become more capable. Believe me i love OSS generally, and i do see all the benefits, but if you're closing your eyes to the risks and only seeing the benefits then you're not doing yourselves any favors. The one thing Dario has said recently that was spot on was that historically the *ability* to do something really bad (think Die Hard 4 or Mr Robot) was negatively correlated with the *desire* to do something really bad. That's no longer the case. I don't know what to do about it. Banning OSS won't work. What's needed maybe is a federal level effort using the best tools to automatically find and fix every Internet exposed vulnerability in every critical system, not once but ongoing. Defenders (and I'm one of them) simply don't have the capability, time, or resources to do this for ourselves. Blame us if you want, but it's the truth: we can't defend against what's coming on our own.

[Jimmy 🇺🇸🇺🇦 @because_sanity]

I make enough money that I could have pretty much any car i want. My first "Tessie" was a 2023 mY RWD. Upgraded to Juniper AWD premium as soon as they were available in my area. I will never own another car that can't drive itself. Once you've experienced it, and truly given into it (it takes s minute), there's no going back to a regular car. I have back and neck issues (5 surgeries, 4 levels fused) and although I've only had one accident that was (arguably) my fault in over 40 years of driving, i still feel safer letting it drive me everywhere. There are 6 million accidents and more than 30k fatalities in the US alone every year. There doesn't need to be. From this point on every human who dies in a car crash is the fault of regulators who are moving too slow to approve FSD unsupervised,

[Jimmy 🇺🇸🇺🇦 @because_sanity]

@RyanCaldwelll @steipete What are you building or trying to accomplish? Also what model(s)? It's almost always faster to just build your own skills now.