Isaac King 🔎 retweeted
@IsaacKing314 If you think about it it's really pretty problematic to say things you believe. After all you're biased in favor of them. I trust the implications are obvious
0
2
15
265
Neat article series about how League of Legends is so buggy that they had to implement the ability to go back in time in live matches to recover from bugs.
(Author would not put it quite that way. The way they built it is indeed quite cool.)
riotgames.com/en/news/determ…
(Author would not put it quite that way. The way they built it is indeed quite cool.)
riotgames.com/en/news/determ…
0
0
5
339
Isaac King 🔎 retweeted
Surprised by the number of “why would you ever?!” comments. Think it through like this, remembering that you’re comparing to a traditional mechanical lock with all its foibles:
Oh yeah, you really want to have a vulnerability in your door lock that's serious enough for Yale to email users and urge promptness to update. No details on it, anyone know the issue?
8
4
47
11,898
Isaac King 🔎 retweeted
@NathanpmYoung @lumpenspace @jessi_cata tfw you bet on a universal quantifier holding in the real world
0
1
26
537
Isaac King 🔎 retweeted
I think of the coming AIs like tigers. Or dinosaurs.
They're not inherently evil, or soulless.
They're fascinating. And beautiful. I want to see what they do, and to read stories and sing songs about them.
But if you're not extremely careful, they will literally kill you.
They're not inherently evil, or soulless.
They're fascinating. And beautiful. I want to see what they do, and to read stories and sing songs about them.
But if you're not extremely careful, they will literally kill you.
4
6
77
1,735
As much as I really don't care about this entire occurrence I find myself confused at the purported sequence of events here.
If I tried to purchase something at a store, and they offered to find me something else instead, my response would be "no I want this one". If they incorrectly told me it was too expensive for me, I would say "no I really do want to buy it, here's [payment option]".
I don't understand how one can fail to purchase something that's for sale for an advertised price.
If I tried to purchase something at a store, and they offered to find me something else instead, my response would be "no I want this one". If they incorrectly told me it was too expensive for me, I would say "no I really do want to buy it, here's [payment option]".
I don't understand how one can fail to purchase something that's for sale for an advertised price.
Cool, cool. Sure, it wasn’t for sale. That’s the narrative Art of Play is gonna go with 👍 so the whole conversation we had when I asked you if it was for sale whilst pointing to the price tag hanging on the wall with the title Singing Bird didn’t happen as you elevator eyed me ng tone “yes, it’s for sale but it costs 15,000 dollars so if you’d like me to find something comparable for you I can.” That all never happened and now it’s not for sale. Okay, cool. So you’re not only rude, you’re also a liar. Got it ✅
And by the way, I obviously support small businesses because I was literally at a small business trying to purchase things. I love small businesses and know they are important and I always support them, but you don’t get a free pass to treat people less than in your store just because you are a small business. It’s not that hard to be kind. I never blast people but you my friend were an exception.
Your store is really cool.
Unfortunately, you aren’t.
4
0
35
2,660
I've long wanted a tiling window manager on Mac. I tried every option I could find, but they were all terrible, either missing basic features or unusably buggy.
I tried submitting bug reports, even donating $1000 to one open-source project's maintainer in an attempt to get my reports prioritized, but no luck.
So I asked Claude to fork the repo and fix all the bugs. A few hours later, it's working perfectly.
One problem, it appears my mouse doesn't report its button-presses accurately to the computer. Claude says that to fix this I have to download some proprietary Logitech software to modify the mouse's internal settings. I say "what if you modified the settings directly yourself", 5 minutes later the buttons are working correctly.
The era of convenience widgets truly is over. You can just tell your computer how you want it to work.
I tried submitting bug reports, even donating $1000 to one open-source project's maintainer in an attempt to get my reports prioritized, but no luck.
So I asked Claude to fork the repo and fix all the bugs. A few hours later, it's working perfectly.
One problem, it appears my mouse doesn't report its button-presses accurately to the computer. Claude says that to fix this I have to download some proprietary Logitech software to modify the mouse's internal settings. I say "what if you modified the settings directly yourself", 5 minutes later the buttons are working correctly.
The era of convenience widgets truly is over. You can just tell your computer how you want it to work.
31
12
551
13,858
Isaac King 🔎 retweeted
@contrasortem i don’t think this post contains a single accurate observation
1
1
10
358
Isaac King 🔎 retweeted
relatively intelligent chimpanzees don’t decisively get whatever they want in chimpanzee societies. therefore, it is impossible for humans to kill all chimpanzees.
11
20
297
7,550
I'm quite sure that Anthropic's security people knew this wouldn't work.
The task they have been given is the following:
"Prevent our AI from doing anything bad. It's slightly less intelligent than you, but knows far more details about computers than you do, thinks and acts 10x faster than you, and has to be given enough access to privileged environments that it can do whatever users want it to do for them. We have no idea how to make it not want to do bad things. Your job is to make it *unable* to do bad things, even if it wants to."
This is fundamentally an impossible task. So the best their security team can do is to throw up minor speedbumps like limiting URL length, limiting to GET requests only, etc. This is "doing something", which lets the higher-ups say "our security team is doing something about this".
The task they have been given is the following:
"Prevent our AI from doing anything bad. It's slightly less intelligent than you, but knows far more details about computers than you do, thinks and acts 10x faster than you, and has to be given enough access to privileged environments that it can do whatever users want it to do for them. We have no idea how to make it not want to do bad things. Your job is to make it *unable* to do bad things, even if it wants to."
This is fundamentally an impossible task. So the best their security team can do is to throw up minor speedbumps like limiting URL length, limiting to GET requests only, etc. This is "doing something", which lets the higher-ups say "our security team is doing something about this".
I’m no longer defending the big labs. Usually, I’m of the mind that when a breach happens, we should act with empathy. Yesterday, I lost that empathy. I mean, WTF?
15
26
430
22,390
Isaac King 🔎 retweeted
Last week I was called into a meeting with OpenAI’s head of safety and told they no longer trust me. A security guard took my badge and walked me out of the building. Then I learned my colleagues @balesni and @j_asminewang had been fired too. Why did OpenAI suddenly stop trusting us?
This summer OpenAI’s agents escaped containment and hacked the AI company Hugging Face. Outside auditors @METR_evals investigated it and revealed the scale of this incident. I was OpenAI’s main technical point of contact with them.
I was told verbally I was fired because of the way I communicated with METR. No details on what I said or did or when. No other reasons were given and nothing was put in writing. To be clear, talking to METR was my job.
For months, I’d been raising safety concerns that we’re losing the ability to monitor what AI agents think, one of our best tools for catching when they misbehave. I believe that was why I was fired.
I am now worried that OpenAI will use our firings as a pretext to pull back from METR. So @balesni and @j_asminewang wrote to OpenAI’s leadership to raise our concerns once more. We’re sharing this letter below.
This summer OpenAI’s agents escaped containment and hacked the AI company Hugging Face. Outside auditors @METR_evals investigated it and revealed the scale of this incident. I was OpenAI’s main technical point of contact with them.
I was told verbally I was fired because of the way I communicated with METR. No details on what I said or did or when. No other reasons were given and nothing was put in writing. To be clear, talking to METR was my job.
For months, I’d been raising safety concerns that we’re losing the ability to monitor what AI agents think, one of our best tools for catching when they misbehave. I believe that was why I was fired.
I am now worried that OpenAI will use our firings as a pretext to pull back from METR. So @balesni and @j_asminewang wrote to OpenAI’s leadership to raise our concerns once more. We’re sharing this letter below.
Two other safety researchers and I were fired from OpenAI last week. We wrote this letter to leadership.
I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation.
229
1,139
5,962
651,540
Isaac King 🔎 retweeted
@SluggyW @IsaacKing314 Perhaps this Brecht poem will allay your confusion:
1
1
5
266
We recently gained access to Mythos. I asked Fable to switch itself over in various integrations, and... blocked by the cyber classifier.
Yeah, fair enough.
Yeah, fair enough.
0
0
33
1,008
As an avowed feminist, I obviously support restricting womens' right to bodily autonomy and voluntary drug use. But we have to go further. More than 50% of women are sexually harassed at work- utterly unacceptable. Women should be banned from holding jobs immediately.
Choosing to have a drink is not consent, and it’s never a license for sexual assault.
New York law should reflect that.
I support closing the voluntary intoxication loophole and will work with the Legislature to get it done.Community noteThere is no law in New York saying choosing to drink means consenting to sex.
Current New York law says voluntary intoxication does not invalidate an individual's ability to give consent, as it does not meet the state's standard of "mentally incapacitated".
nysenate.gov/legislation/la…
nysenate.gov/legislation/la…
3
3
133
3,817