Max Talwar

Economics + Math @ Georgetown

[Max Talwar @max_talwar1]

Here's an even more generalizable counterexample, which extends to an infinite family. What a time to be alive!

[Matty Hempstead @mattyhempstead]

WOWII Conjecture 91 is false. This graph theory problem was open for ~22 years.

I cracked it by asking ChatGPT 5.6 Pro to find a counterexample to an open conjecture of its choosing and then I went to have a nap.

I do NOT care for graph theory nor had I heard of this NERD conjecture until ChatGPT told me it found a solution.

ChatGPT conversation where this was found:
https://t.co/4AnMhde026

[Hunter📈🌈📊 @StatisticUrban]

Consider that the median person, if they know anything about AI at all, still thinks it's a basically a video slop generator that can maybe cheat on your homework for you.

Very few have any idea what's coming.

[Lomez @L0m3z]

Probably my mostly deeply held liberal view is that America is uniquely successful at creating upward mobility and vasts sums of wealth and opportunity for people who otherwise would be limited by inequalities that have nothing to do with personal merit, and that our system of higher-ed was one of the main engines for this world historical triumph of personal advancement, perhaps *the* shining example of our civilizational might, and that it is among our greatest shames and most tragic failures that we have allowed the institution to kill itself on the alter of the most stupid and trivial political hysterias, one after the other, decade after decade, with no end in sight and no strategy or plan at all for how to fix it.

[The Ramen Don @RealRiddimHours]

How many of these fucking hearings do we have to have where the top BRASS is saying "yeah the Ukrainians are lowkey goated w the sauce idk what to tell you, and they love killing our enemies" before it becomes obvious to these fucking people what to do next

[Max Talwar @max_talwar1]

@gakonst We don't, but maybe we should!

[Max Talwar @max_talwar1]

What's so amazing about this is that the author just took a conjecture, told 5.6 to find a counterexample, and every time it said it couldn't find one he just said "try again" until it eventually did.

Math is in the same place as vibecoding a year ago!

[Dmitry Rybin @DmitryRybin1]

Dinitz-Garg-Goemans conjecture is false. This graph theory problem was open for ~30 years.

The graph below has fractional flow cost 58. Any unsplittable flow (with capacity violation <=15) has cost at least 60.

Chat with GPT 5.6 Pro where this was found: https://t.co/Oi2PQoab2h

[Max Talwar @max_talwar1]

@perrymetzger Yeah, both labs have been trying REALLY hard to show everyone how dangerous their models are, and this certainly helps OpenAI's case.

[Max Talwar @max_talwar1]

@cremieuxrecueil Well I think the issue is that it's a very slippery slope to have someone / some organization become the arbiter of truth, as tempting as it may seem.

That's how you become the UK!

[Max Talwar @max_talwar1]

@ChunChiuLu @Jason The overwhelming majority of our deficit comes from lavish entitlement programs, something DOGE isn't going to fix.

Also he controlled essentially every branch of government. He stopped himself!

[Max Talwar @max_talwar1]

@Iovesnakee @alz_zyd_ I don't think anyone is arguing that it's a cornerstone of Marxian economics, rather that it's pretty reflective of his quantitative skills, and not in a good way!

[Steve Hou @stevehou]

Unpopular opinion: AI basically became useful for enterprise use early this year with agentic AI acceleration. Everything before was basically consumer and hobbyist. 2026 is year 0 for AI enterprise use. So why would anyone have expected measurable productivity increase already.

[Max Talwar @max_talwar1]

@djego0101010 @alz_zyd_ And doing a pretty poor job of it :)

[Max Talwar @max_talwar1]

@LeeKuanYimby My goat is back

[Max Talwar @max_talwar1]

@AravSrinivas Fair point, but have you see how immensely expensive running a model like K3 is locally? The practical local models are far behind the frontier, and by definition probably will continue to be because running locally requires less compute for inference.

[Nate @natehoodstp]

It’s pretty simple:

American society has prioritized private space. We have, on average, better, larger, and nicer homes

European society has prioritized public space. They have, on average, better, larger, and nicer parks, plaza, transit, shopping districts, etc.

[Max Talwar @max_talwar1]

@alz_zyd_ I think this little "proof" of his is actually a great window into how Marxism as a whole works. Even the fact that he just assumed dy and dx are literally zero is so perfect, he literally has no consideration for marginal effects if they look small enough!

[Max Talwar @max_talwar1]

I suspect marginal GPU increase is too small. The compute, RAM, etc from any individual device is so tiny that you would have to wire together huge numbers of devices from around the world to do a single task, which is so much less efficient than a data center.

Most compute from these consumer devices are also not specialized for AI the way ASICs/most Nvidia chips are.

[Max Talwar @max_talwar1]

@sigmahamster2 Yeah Lee Kuan Yew was awesome but like the opposite of a libertarian. He was a very effective technocrat!

[Max Talwar @max_talwar1]

@hustle_fred @levelsio I think you should do some deep thought about what the potential second order effects of allowing any inconvenient truth to be called defamation might be

[Max Talwar @max_talwar1]

@Laproxen @AbakpaJob The image is a metaphor, hope this helps!