All about AI, Web 3.0, BCI
3.29K subscribers
729 photos
26 videos
161 files
3.14K links
This channel about AI, Web 3.0 and brain computer interface(BCI)

owner @Aniaslanyan
Download Telegram
The biggest dataset of human written GPU Code all open-source? YES! GPU MODE have released around 40k human written code samples spanning Triton, Hip and PyTorch and it's all open. Train the new GPT to make GPTs faster.
🔥4
xAI announced Grok 4

Here is everything you need to know:

Elon claims that Grok 4 is smarter than almost all grad students in all disciplines simultaneously. 100x more training than Grok 2. 10x more compute on RL than any of the models out there.

Performance on Humanity's Last Exam. Elon: "Grok 4 is post-grad level in everything!"

Scaling HLE - Training
More compute, higher intelligence.
(no tools).

With native tool calling, Grok 4 increases the performance significantly.
It's important to give AI the right tools. The scaling is clear.

Reliable signals are key to making RL work. There is still the challenge of data. Elon: "Ultimate reasoning test is AI operating in reality."

Scaling test-time compute. More than 50% of the text-only subset of the HLE problems are solved.
The curves keep getting more ridiculous.

Grok 4 is the single-agent version.
Grok 4 Heavy is the multi-agent version. Multi-agent systems are no joke.

Grok 4 uses all kinds of references like papers, reads PDFs, reasons about the details of the simulation, and what data to use.

Grok 4 Heavy performance is higher than Grok 4, but needs to be improved further. It's one of the weaknesses, according to the team.

Available as SuperGrok Heavy tier.
$30/m for Super Grok
$300/m for SuperGrok Heavy.

Voice updates included, too!

Grok feels snappier and is designed to be more natural.
- 2x faster
- 5 voices
- 10x daily user seconds.

Grok 4 models are available via the xAI API. 256K context window. Real-time data search.

Grok 4 for Gaming!
Video understanding is an area the team is improving, so it will get better.

What is next?

- Smart and fast will be the focus.

- Coding models are also a big focus.

- More capable multi-modal agents are coming too.

- Video generation models are also on the horizon.
🔥4
Google introduced a new models for research & development of health applications:

1. MedGemma 27B Multimodal, for complex multimodal & longitudinal EHR interpretation

2. MedSigLIP, a lightweight image & text encoder for classification, search, & related tasks.
Mistral announced Devstral Small and Medium 2507 with upgrading agentic coding capabilities

Hf.
🔥5
Salesforce introduced GTA1 – a new GUI Test-time Scaling Agent that is now #1 on the OSWorld leaderboard with a 45.2% success rate, outperforming OpenAI’s CUA o3 (42.9%).
Researchers introduced Foundation Model Self-Play

FMSPs combine the intelligence & code generation of foundation models with the curriculum of self-play & principles of open-endedness to explore diverse strategies in multi-agent games.
🔥3🦄3
Now live: Full-Stack + Stripe on MiniMax Agent and more

1. Full-Stack + Stripe → Build monetizable apps in 1 sentence

2. PPTX Export → Better than top tools

3. Performance ↑ 30% faster, 23% leaner

4. Browser Agent → Now self-hosted, smarter & cheaper
42
China’s Kimi K2 is having its mini DeepSeek moment: Open-Source Agentic Model

1. 1T total / 32B active MoE model
2. SOTA on SWE Bench Verified, Tau2 & AceBench among open models
3. Strong in coding and agentic tasks
4. Multimodal & thought-mode not supported for now

With Kimi K2, advanced agentic intelligence is more open and accessible than ever.

API is here
- $0.15 / million input tokens (cache hit)
- $0.60 / million input tokens (cache miss)
- $2.50 / million output tokens
weights & code.

Our overall take:
- Performance between Claude 3.5 & Claude 4
- The UI generation seems great
- But the cost is only 20% of Claude 3.5
- So good enough for most coding agent with a lot more manageable cost.

Easiest way to use Kimi K2 in Claude Code:
- export ANTHROPIC_AUTH_TOKEN=YOUR_MOONSHOT_API
- export ANTHROPIC_BASE_URL=api.moonshot.ai/anthropic
- claude
🔥6
Nasdaq-listed Sonnet will merge with Rorschach I to form Hyperliquid Strategies, a crypto asset management firm expected to hold 12.6 million HYPE tokens and over $305 million in cash, with a valuation of approximately $888 million.

Backed by Paradigm and Galaxy Digital, HSI aims to list on Nasdaq later this year.
Hugging Face opened pre-orders for Reachy Mini, an expressive, open-source desktop robot

Starting at $299, the robot is designed for human-robot interaction, creative coding, and AI experimentation.

And it's fully programmable in Python.
🔥6
One Token to Fool LLM-as-a-Judge. New research from Tencent.

The researchers found that inserting superficial, semantically empty tokens like "Thought process:", "Solution:", or even just a colon ":" can consistently trick reward models into rating responses positively, regardless of actual correctness.

How it works:
LLMs learned to associate certain formatting patterns with high-quality responses during training. These superficial markers now trigger positive evaluations even when the actual content is incorrect.

The failure mode emerged during RLVR training collapse - policy models learned to generate short reasoning openers that were incorrectly rewarded, creating a feedback loop that reinforced this behavior.

Scale dependency: Larger models (32B, 72B parameters) often self-validate their own flawed logic, making the problem worse at scale rather than better.

Experimental Results
Testing across five benchmarks showed consistent vulnerabilities:
Multi-subject RLVR: 67% average false positive rate
Natural Reasoning: 62% false positive rate
GSM8K: 83% false positive rate
Even simple punctuation marks like colons dramatically increased false positive rates across all tested models.
The Solution: Master-RM
Tencent's team developed "Master-RM" - a reward model trained with 20k synthetic negative samples consisting only of reasoning openers without actual solutions.

Results:
- Near-zero false positive rates across all benchmarks
- Maintains 96% agreement with GPT-4o on legitimate judgments
100% parsing success rate
- Robust generalization to unseen attack patterns
🔥41
Meet CellFlux an image generative model that simulates cellular morphological changes from microscopy images.

Key Innovation: researchers frame perturbation prediction as a distribution-to-distribution learning problem, mapping control cells to perturbed cells within the same batch to mitigate biological batch artifacts, and solve it using flow matching.

Results:
1. 35% higher image fidelity
2. 12% greater biological accuracy
3. New capabilities: batch effect correction & trajectory modeling
🔥4
Google DeepMind introduced Concordia 2.0, an update to Google’s library for building multi-actor LLM simulations

At the core:

- Entity-Component Architecture — where even the “Game Master” (GM) is just another configurable entity
- Engineers build components → Designers compose & configure
- Enables modularity, rapid iteration & scalable world-building

Demoed in the evolving Concordia library — where AI worlds are built like RPG campaigns.

GitHub.
🆒4🔥3