Yacine Mahdid

@yacinelearning

(neuro/ai) I make technical deep learning tutorials 👺

following
followers

VPO post-training aims to produce diverse sets of good AI answers

A participant says a two-hour discussion covers why diversity matters and reward shaping for exploration.

4 Sources, 5d ago, first seen 5d ago

Capture proxy enables reinforcement learning for open models across coding-agent harnesses

A post says the OpenEnv proxy records token IDs and log probabilities while harness runs remain unmodified.

22 Sources, 14d ago, updated 14d ago

Arcee AI announces Series B funding at a valuation above $1 billion

Arcee says the funding will advance its Trinity models, expand work with the Department of Energy and national labs, and support a platform for building and running open models.

11 Sources, 34d ago, first seen 34d ago

Hutter Prize challenge announcement invites an AI “agent swarm” to tackle compression

The invitation asks participants to give their agent a write token for a Hugging Face organization, then paste a setup snippet into Codex, Claude Code or Hermes Agent.

6 Sources, 38d ago, first seen 38d ago

Y Combinator says a better harness lifts the same AI model from 30% to 95% on ARC-AGI

YC says it gathered researchers and founders to explore AI harnesses—the software around models—and lessons from building an agent for every employee at the company.

8 Sources, 43d ago, first seen 43d ago

Interview on Prime Agent Harness Announced

Yacinelearning announces interview with Seth Karten tomorrow at 14h00 EST.

2 Sources, 48d ago, first seen 48d ago

Mithil Vakde Reaches 44% on ARC-AGI-1 Benchmark

Yacine Mahdid posted about efficient training of a compact transformer model.

1 Source, 49d ago, first seen 49d ago

Recirculation Technique Recycles Activations in Transformers

The method reuses top-layer activations in lower layers during inference on existing models.

18 Sources, 62d ago, first seen 62d ago

Featured in

VPO post-training aims to produce diverse sets of good AI answers

Capture proxy enables reinforcement learning for open models across coding-agent harnesses

Arcee AI announces Series B funding at a valuation above $1 billion

Hutter Prize challenge announcement invites an AI “agent swarm” to tackle compression

Y Combinator says a better harness lifts the same AI model from 30% to 95% on ARC-AGI

Interview on Prime Agent Harness Announced

Mithil Vakde Reaches 44% on ARC-AGI-1 Benchmark

Recirculation Technique Recycles Activations in Transformers

Featured in

VPO post-training aims to produce diverse sets of good AI answers

Capture proxy enables reinforcement learning for open models across coding-agent harnesses

Arcee AI announces Series B funding at a valuation above $1 billion

Hutter Prize challenge announcement invites an AI “agent swarm” to tackle compression

Y Combinator says a better harness lifts the same AI model from 30% to 95% on ARC-AGI

Interview on Prime Agent Harness Announced

Mithil Vakde Reaches 44% on ARC-AGI-1 Benchmark

Recirculation Technique Recycles Activations in Transformers