Cognition's Devin agent resolves decades-old open math conjectures

WDwill depue@willdepue1h ago · 32.3K Views

366TECH

guys you can’t claim your Devin solved unsolved problems when it’s literally Fable and 5.6 calls in the backend. come on

JZ

Jared Zoneraich@imjaredz

Proud to say that Devin has cracked three more unsolved problems today 1) REFUTED: Graffiti Conjecture 154 (open for ~40 years) 2) PROVED: Graffiti Conjectures 39 & 40 (~40 years) 3) REFUTED: Brandt's Regular Supergraph Problem from West's open problems list (~20 years) My methodology is explained below, but basically I showed Devin the original tweet and told it to find similar problems and crack them. The flood gates are open.

LALisan al Gaib@scaling013h ago · 19.1K Views

1150TECH

the math unhobbling is happening

JZ

Jared Zoneraich@imjaredz

Proud to say that Devin has cracked three more unsolved problems today 1) REFUTED: Graffiti Conjecture 154 (open for ~40 years) 2) PROVED: Graffiti Conjectures 39 & 40 (~40 years) 3) REFUTED: Brandt's Regular Supergraph Problem from West's open problems list (~20 years) My methodology is explained below, but basically I showed Devin the original tweet and told it to find similar problems and crack them. The flood gates are open.

JLJoe Lonsdale@JTLonsdale3h ago · 14.2K Views

758TECH

Cool. Everything is about to accelerate… age of wonders if we get it right.

JZ

Jared Zoneraich@imjaredz

Proud to say that Devin has cracked three more unsolved problems today 1) REFUTED: Graffiti Conjecture 154 (open for ~40 years) 2) PROVED: Graffiti Conjectures 39 & 40 (~40 years) 3) REFUTED: Brandt's Regular Supergraph Problem from West's open problems list (~20 years) My methodology is explained below, but basically I showed Devin the original tweet and told it to find similar problems and crack them. The flood gates are open.

YOYohei@yoheinakajima3h ago · 4.6K Views

1077TECH

it’s like a world wide easter egg hunt right now

JZ

Jared Zoneraich@imjaredz

Proud to say that Devin has cracked three more unsolved problems today 1) REFUTED: Graffiti Conjecture 154 (open for ~40 years) 2) PROVED: Graffiti Conjectures 39 & 40 (~40 years) 3) REFUTED: Brandt's Regular Supergraph Problem from West's open problems list (~20 years) My methodology is explained below, but basically I showed Devin the original tweet and told it to find similar problems and crack them. The flood gates are open.

CSChristian Szegedy@ChrSzegedy1h ago · 2.1K Views

105TECH

Just a drizzle before the storm.

RPRohan Paul@rohanpaul_ai1h ago · 2.2K Views

1252TECH

Another Math win for AI. 2 conjectures from the 80s and one from 2002 all died this week. and this is the slowest AI math will ever be. Any problem with a checkable answer can be potentially solved at the speed of inference. Mathematical knowledge has always grown at the pace of human patience, but now that has been replaced with compute, and compute compounds.

JZ

Jared Zoneraich@imjaredz

Proud to say that Devin has cracked three more unsolved problems today 1) REFUTED: Graffiti Conjecture 154 (open for ~40 years) 2) PROVED: Graffiti Conjectures 39 & 40 (~40 years) 3) REFUTED: Brandt's Regular Supergraph Problem from West's open problems list (~20 years) My methodology is explained below, but basically I showed Devin the original tweet and told it to find similar problems and crack them. The flood gates are open.

JZJared Zoneraich@imjaredz5h ago · 1.3K Views

I got totally nerd sniped today I'll share more of the methodology soon, but basically I had a few Devins research good problems similar to the one the OP solved Then I picked the best problems Then I spun up ~5 Devins in Ultra mode for each problem to find & verify solutions (machine-checked in Lean) Repo here https://github.com/jzone3/solving-math-problems

WDwill depue@willdepue1h ago · 230 Views

@imjaredz

ACAndrew Curran@AndrewCurran_3h ago · 2.9K Views

618TECH

Devin.

JZ

Jared Zoneraich@imjaredz

Proud to say that Devin has cracked three more unsolved problems today 1) REFUTED: Graffiti Conjecture 154 (open for ~40 years) 2) PROVED: Graffiti Conjectures 39 & 40 (~40 years) 3) REFUTED: Brandt's Regular Supergraph Problem from West's open problems list (~20 years) My methodology is explained below, but basically I showed Devin the original tweet and told it to find similar problems and crack them. The flood gates are open.

JZJared Zoneraich@imjaredz3h ago · 604 Views

EDIT: Looks like #154 was solved a month ago in https://github.com/demonstrandum-research/artifacts/blob/main/RESULTS.md

JZJared Zoneraich@imjaredz4h ago · 661 Views

just solved another unsolved problem btw

JZJared Zoneraich@imjaredz5h ago · 1.4K Views

1) Graffiti Conjecture 154 REFUTED. Open for ~40 years. Devin found the counterexample: glue a 50-clique to a 70-edge path (n=120). The violation reduces to a single integer inequality, proven in Lean down to the graph's distance sum. 2) Graffiti Conjectures 39 & 40 PROVEN. From the same 1988 Graffiti program. The standard deviation of a graph's distance matrix never exceeds its number of positive (#39) or negative (#40) adjacency eigenvalues. Both are TRUE. It’s a 4-line proof chaining Popoviciu's inequality on variances with a 2020 spectral lemma (Geng–Wu–Wang) nobody had connected to them. 3) Brandt's Regular Supergraph Problem from West's open problems list REFUTED. Found a 9-vertex counterexample where no amount of vertex cloning (duplicating a vertex with the same neighbors) can ever make it regular. Certified by an exact Farkas infeasibility certificate Now it's possible this is actually a 20-year-old typo. Brandt's original paper says strictly greater than n/3, but West's open problems list says at least n/3. And it's false exactly at that boundary.

BRBlue Rose@latereigns5h ago · 680 Views

@imjaredz All attributions shall go to Soltra, let's go.

JZJared Zoneraich@imjaredz1h ago · 375 Views

@willdepue I was able to solve this precisely because of the Devin harness and the parallelization it gives me out of the box Fable and 5.6 are excellent models and it’s obvious are the main drivers but I just wouldn’t have gotten these results so fast with a single cli agent

BEbellatrix@0xb3llatrix3h ago · 138 Views

@imjaredz @tweem likes 100k

ARar0cket1@ar0cket13h ago · 126 Views

@imjaredz is that fable, 5.6 sol?

BEBen@BensBeardButter2h ago · 22 Views

@JTLonsdale Hey Joe how about instead of giving 90% of your wealth to taxes you pull together something considerable and drive a call to action amongst homies to run back the Rough Riders bringing in Baja California as a new economic zone while tapping into equity @LatinoFreeMinds for @GOP

WDwill depue@willdepue1h ago · 106 Views

@imjaredz all in good fun, i think you get the point

JLJoe Lonsdale@JTLonsdale2h ago · 85 Views

@BensBeardButter @LatinoFreeMinds @GOP I am doing very bold stuff like this except the IQ-160 version with a lot of talented and aligned people in my policy group, university, and with mission-driven companies. I agree we need to be courageous and smart, and fight for our country and our future in novel ways.

BEBen@BensBeardButter2h ago · 14 Views

@JTLonsdale @LatinoFreeMinds @GOP That's true you are doing great work with Cicero and Univesity of Austin I don't mean to dismiss - but Rough Riders has layers that integrate in terms of getting into the liver of progressive policy with prison reform that you're being sub 160 IQ on to dismiss so easily