Isaac Was Here (@agnoster) profile picture

Moved on for bluer skies (@agnoster.net)

Not Twitter anymore
Joined January 2007
Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)Photo by Isaac Was Here (@agnoster)
Alrighty, you know what? That's my cue. I'm "agnoster" on 🧵 and "agnoster.net" on 🔵☀️ See you there, or IRL. I'm out.
1
1
906
Ok this was worth dusting off my Twitter login credentials to share
159
"Elon YOLO'd some prompt changes in hoping to make Grok believe his white genocide bullshit. We can't say it was Elon because he's technically the boss, but, cmon, who has high level access and is obsessing over 'Kill the Boer' at 3am, you guys do the math. This dude suuuuuucks"
We want to update you on an incident that happened with our Grok response bot on X yesterday. What happened: On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot's prompt on X. This change, which directed Grok to provide a specific response on a political topic, violated xAI's internal policies and core values. We have conducted a thorough investigation and are implementing measures to enhance Grok's transparency and reliability. What we’re going to do next: - Starting now, we are publishing our Grok system prompts openly on GitHub. The public will be able to review them and give feedback to every prompt change that we make to Grok. We hope this can help strengthen your trust in Grok as a truth-seeking AI. - Our existing code review process for prompt changes was circumvented in this incident. We will put in place additional checks and measures to ensure that xAI employees can't modify the prompt without review. - We’re putting in place a 24/7 monitoring team to respond to incidents with Grok’s answers that are not caught by automated systems, so we can respond faster if all other measures fail.
87
Been wanting something more like this for all the Paradox games tbh - let me control high level levers and see how the *systems* behave based off the incentives/constraints you set up. More sim, less micro.
i wish cities skylines had an autonomous mode where new citizens would figure out on their own what there is demand to build and anything is permitted unless you intervene by adding zoning or other policies
1
379
Isaac Was Here retweeted
there are really only two types of people: 1. got mine, fuck you 2. nobody should have to endure this the rest is just details
115
3,508
185
10,378
You've heard "LLMs are calculators for words" (simonwillison.net/2023/Apr/2…) We already had calculators for words 10 years ago (word2vec) - but pocket calculators started with numbers and then moved to functions, then programs I think LLMs are now closer to "calculators for ideas"
147
I see a lot of "give a good argument for AI risk that's simple and doesn't have external reading" I think that's a great challenge, but we should also remember there's no reason the universe needs to shape existential risks in a way that lends itself to pithy explanation
1
2
1,170
If, counter-factually, the LHC had had some kind of x-risk, but to understand it you'd need to know quantum mechanics, it wouldn't have been less real We knew the Manhattan project wouldn't start a runaway chain reaction in atmospheric nitrogen, but it was a lot of complex math
1
2
1,049
Of course we should try to distill arguments in as clear a way as possible, but we should also have enough epistemic humility to know that between a complex, nuanced argument and a short pithy sound-bite, truth is not a question of brevity.
1
464
Today I was doing an author talk at a middle school when a lockdown was called. I asked the girl nearest to me what to do. I explained that I'd never done a lockdown before. She looked at me in disbelief and said, "Did you grow up in America?"
182
1,685
245
16,368
1,568,340
Python: there should be one and preferably only one obvious way to do it Also python: pyenv virtualenv venv anaconda miniconda pip setuptools easy_install wheel poetry requirements.txt pyproject.toml idk you figure it out
The Python noob experience: mention the "cognitive overload of having 20 different ways to install things", get a dismissive answer that touches on virtualenv, pip, setuptools, wheel...
3
796
Isaac Was Here retweeted
Hi! Please let LLMs do your code review. It's really good at it and catches things your team probably won't. I made a little a tool you can use to do this: lintrule.com, hooks right into your CI. Write rules in english and run "rules check", like a test framework
45
81
20
780
186,680
PSA: ChatGPT Browsing and Plugins should now be available to all Plus subscribers (rollout complete). 👏 Make sure to enable these new capabilities in the settings via "Beta features". Enjoy! 🔌
56
32
15
425
120,391
Still thinking about this piece from @micahdaigle pretty regularly tbh I’m a sucker for tools for explicitly switching cognitive “modes” micahdaigle.com/thoughts/ros…
1
1
143
Isaac Was Here retweeted
Just launched the ChatGPT iOS app! apps.apple.com/us/app/openai… Now in the US, world soon. Android next.
444
1,263
357
7,691
2,810,933
My #1 critique of moral systems that require scoring people instead of behaviors and actions. If you can't acknowledge bad choices you've made without collapsing your self worth, you can never really be morally honest with yourself, and thus never improve.
It seems like a good principle to hold that "If you know the right thing to do, but choose to do something else because it's easier, you're a bad person", but in practice this likely just shunts you into self deception about what's the right thing to do
1
158