Kevin Lacker

Working on math + AI at and telescope software at Formerly: Parse cofounder, Facebook, Google

[Kevin Lacker @lacker]

I'm happy to announce the launch of Acorn, a new theorem prover that includes an integrated AI.

Theorem provers let you write mathematical proofs that are rigorously verified. But they are notoriously difficult to use. Acorn makes it easier, by using AI to fill in the details.

[Kevin Lacker @lacker]

@JI Yeah, I believe it. The LLMs seem like a super tempting way to trick yourself into believing you understand things better than you do. Or just to delay an inevitable F rather than coming to terms with it early and changing course.

[Kevin Lacker @lacker]

Entertained by professor Garcia here.

1. Claims he is a "strong, strong opponent" of curving a class so that only a limited number of students can get A's

2. Fails 35% of the students in CS 10, saying they all deserve to fail

https://t.co/SSmFEZ854o

[Kevin Lacker @lacker]

@tonylfeng personally I find these announcements very useful, it's nice to know what the capabilities of the models are!

[Kevin Lacker @lacker]

@gerardsans

[Kevin Lacker @lacker]

I liked this piece... but I felt like it dodged the most interesting questions, by focusing on "is AI conscious".

What would it be like if AI became better at writing than humans?

I'll admit, it's quite hard for me to imagine that world. But I think Ted Chiang could imagine it.

[Adrienne LaFrance @AdrienneLaF]

You really have to read Ted Chiang on consciousness and AI—just published in The Atlantic:

https://t.co/z9eRSlkwQ8

[Kevin Lacker @lacker]

@thomasfbloom this would be a really fantastic outcome, right? so why worry so much about, we must make sure that AI and human mathematics are aligned. isn't it more important to try to make these marvelous things happen in the first place, regardless of the method?

[Kevin Lacker @lacker]

@IlinVasily29521 It looks pretty neat! I wonder, how do you plan on updating Lean versions? Just have an LLM do it for everything at once, and figure that will probably work?

[Kevin Lacker @lacker]

@jamesjyu in Oakland there’s Dumpling Hours

[Kevin Lacker @lacker]

@tlbtlbtlb @rsdgpt Yeah, I have a lot of experience of that from Parse 😜 If you want to be really safe, you can use a read-only replica, but for smaller teams it’s probably fine.

[Kevin Lacker @lacker]

@Akbar501 yep

[Kevin Lacker @lacker]

@samth @QiaochuYuan What if an AI invents some new mathematics and there is no human "within the mathematical community" who deserves any credit for it?

[Kevin Lacker @lacker]

@jamesjyu @ninabegus is there going to be a kindle version?

[Kevin Lacker @lacker]

@hkarthik Tough when California is 17% Asian, the UC system is 50% Asian, and so much of California politics is race-based. Purely anecdotally based on observing groups of kids, I wonder if the politics will change next gen due to booming multi-ethnic population

[Kevin Lacker @lacker]

A really interesting document. My takeaways:

1. the UC student body is about:

20% Hispanic
50% Asian
25% white
5% other

2. Despite all the arguments, laws, and rule changes, that only changed a few % over the past 10 years

[Jelani Nelson @minilek]

One motivation to ban SAT/ACT in UC admissions was to eliminate perceived bias against black and latino students.

But the biggest shift by banning the SAT was a transfer of admit seats from Asian students to white: https://t.co/MUDsL2CSk1

E->F, rows 12-19.

[Kevin Lacker @lacker]

@hkarthik See: https://t.co/HtlEHOR6bW Section 3.1.2: black 4-year graduation rate is 60% compared to asian 80%. Also 3.1.6 has stats for transfers, which are 65% to graduate on time. So, I think the community college route is actually *less* merit based, more political.

[Kevin Lacker @lacker]

@growing_daniel seems like every remaining controller has a limp, wobbly joystick

[Kevin Lacker @lacker]

@Vjeux interesting. i want to try having it build a whole game engine - like a cli/json version of the game. then it’s entirely separate challenges, one is building a faithful model, two is playing

[Kevin Lacker @lacker]

@tonylfeng we need a data set the AI people can train on, like 100 papers that should be accepted and 100 that have subtle errors in them