Bargav Jagatha

[adaption @adaption_ai]

Last week we introduced Invent a Dataset. Describe what you need. Hit go.

Today, we’re making it even easier with the Invent API.

A few lines of code. AI ready training datasets in minutes.

[adaption @adaption_ai]

Bringing AutoScientist to 15+ million @huggingface researchers and builders. 🤗

AutoScientist automates model training. Specify your objective, let AutoScientist do the rest.

Export to Hugging Face with one click.

[Sudip Roy @sudip_r0y]

Today we’re launching Invent a Dataset at Adaption.

Describe what you want a model to learn. Invent a Dataset generates the training data; AutoScientist turns it into an optimized model.

From specification to dataset to an owned model.

[Bargav Jagatha @bargav025]

no seeds. no scraping. describe it, get the dataset. lfg 🚀

[adaption @adaption_ai]

Introducing Invent a Dataset.

Describe the dataset you need. Get a structured, training-ready dataset back. No existing data needed.

Dataset creation used to start with collection. Now it starts with specification.

[adaption @adaption_ai]

Introducing Invent a Dataset.

Describe the dataset you need. Get a structured, training-ready dataset back. No existing data needed.

Dataset creation used to start with collection. Now it starts with specification.

[Bargav Jagatha @bargav025]

our office is finally officing. yayyy

[Sara Hooker @sarahookr]

We have furnitureeee now.

A whole new world @adaption_ai

[Bargav Jagatha @bargav025]

@bjamin_ar @sarahookr Poor Dhruv had to come up with Thailand story in the spot 😂

[Bargav Jagatha @bargav025]

treating the compiler harness as an evolving interface between agents and rapidly changing gpu hardware.
--- seems killer 🔥

[Zihao Ye @ye_combinator]

🍰 CAKE paper's out, the design bet: the compiler isn't a fixed black box the agent calls — it's part of the harness, and it's under evolution too.

CAKE didn't inherit existing abstraction layer. no tile/layout abstractions: the vocabulary was distilled by agents from a corpus of production kernels. every pattern the agent couldn't express pushed new primitives into the IR, and the analyses to keep them checkable. every barrier/layout bug that kept coming back became a verifier rule.

none of this can be designed up front. the IR has to co-evolve with the kernels, and the workload tells you what's missing, the corpus tells you if the fix broke anything.

the best language for an agent is the one that tells you what's illegal, what's slow, and which decision might made it faster.

https://t.co/810wbeLuAW

[adaption @adaption_ai]

Can the Adaption team guess a legendary researcher from clues alone?

Turns out, yes. Mostly.

@sudip_r0y and the team play Guess the Researcher.

[Bargav Jagatha @bargav025]

Results:

LongMemEval: 90.6%
LoCoMo: 88.2%
BEAM (1M): 61.3%

all these consuming only 4k-6k tokens per query

[Bargav Jagatha @bargav025]

The biggest RAG mistake isn't bad retrieval.

It's bad writes.

Here's what building our own agent memory actually taught us

[adaption @adaption_ai]

The cost of remembering should not scale with how much there is to remember.

New research from @bargav025 and @sudip_r0y: 45% faster on long conversations, 97% fewer tokens processed.

[Bargav Jagatha @bargav025]

But ADD-only creates contradictions.

"Prefers remote work" → "just accepted an in-office role"

Fix: supersession chains -- soft-rank older memories down when newer ones clearly override. Helped a lot on knowledge-update and preference questions specifically.