Finetuning Singh

I like post training models

[Finetuning Singh @FinetuningSingh]

Vibed @typesafeai's Allie Laabs's IntelliPrompter with Jev: a teleprompter of talking points, not a script. Talk freely, in any order, and each point checks itself off once you've covered it.

One Jev Score question per point, every half second, <$0.50/hr.

Try it: https://t.co/2zIbkD3Wgy

[Finetuning Singh @FinetuningSingh]

@typesafeai you can use it here https://t.co/7ybSQBrQA9

[Finetuning Singh @FinetuningSingh]

Turned @typesafeai’s Jev into a next-word predictor over 30k words.

It can’t generate text. For every word, it picks 1 of 254 meaning groups, then picks the word.

Sometimes it’s weirdly right:

“what day comes after monday” → tuesday
“how many legs does a spider have” → viii
“is the earth flat” → no
“i am sad” → sorry you sad understand

The fun part is where it breaks: the answer is often in there, but the sentence falls apart.

“what is the biggest animal” → is the is im is whale
“who wrote hamlet” → is was william
“tell me a joke” → sure tell joke chicken
“what is 2 plus 2” → two

Jev picks. Jev does not write.

https://t.co/wyS05ffo46

[Finetuning Singh @FinetuningSingh]

OpenAI’s Navier–Stokes effort is one of the clearest examples of what AI research orchestration could look like. Instead of one model following one approach, multiple AI agents can explore different mathematical objectives, share useful ideas, and shift compute toward the most promising paths.

[Finetuning Singh @FinetuningSingh]

OpenAI released an AI-generated solution to the Navier–Stokes Millennium Prize Problem, arguing that a smooth 3D fluid can develop a singularity—or “blow up”—in finite time. If the proof survives expert review and the formal acceptance process, it could resolve one of mathematics’ most famous open problems and qualify for the $1 million Millennium Prize. Here’s what the result actually means in under a minute. OpenAI published the result on September 8, 2026, along with a formalized Lean proof.
#OpenAI #NavierStokes #Mathematics #AI #MillenniumPrize #FluidDynamics #Science #Math

[Finetuning Singh @FinetuningSingh]

Mathematics is gambling.

You have an idea, spend hours—or even weeks—trying to prove it, and eventually have to decide whether the path is worth pursuing. But AI has a very different constraint: it can make a mistake, backtrack, abandon a bad approach, and simply try again.

That persistence could become one of AI’s biggest advantages in mathematics. Not because it never fails—but because it can afford to fail far more often than we can.

#AI #Mathematics #Math #ArtificialIntelligence #MachineLearning #AIMath #Shorts

[Finetuning Singh @FinetuningSingh]

High beating XHigh is kind of funny. More reasoning won't always help. #AI #LLM #AIResearch

[Finetuning Singh @FinetuningSingh]

AI has created a new kind of flex: “Look how much I spend on compute.” #futureofwork #gpt6 #ai

[Finetuning Singh @FinetuningSingh]

$50 = ~1 million AI output tokens.
$50 = ~7 hours of minimum-wage labor.
It’s: can 1M tokens produce more useful work than a person can in those 7 hours?