 Your new post is loading...
Your new post is loading...
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
Integrating robust safety guardrails into Large Language Models (LLMs) is essential for delivering helpful yet harmless responses. While proprietary systems exhibit reliable safety controls, their underlying methodologies and trade-offs remain largely undisclosed. Achieving comparable security in open-weight models remains a persistent challenge, as post-trained variants frequently suffer from over-refusal and degraded general quality. To overcome these drawbacks, we introduce Suan, a novel preference optimization algorithm. Unlike existing methods, we formulate the optimization objective directly at the gradient level, bypassing the standard variational derivation. As a result, we obtain more interpretable and robust training dynamics. Extensive evaluations across a diverse suite of competitive baselines and benchmarks demonstrate that Suan achieves superior safety alignment while fully preserving response utility.
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
You can now use the Perplexity Search API in Hermes Agent
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
If you're expecting Nex-N2.5 PRO API
Coming soon 👇 @OpenRouter
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
Cognition's valuation multiple is higher than Cursor's was before selling to SpaceX.
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
Astra is fully rolled out to Plus, Pro, Business, and Enterprise users in Codex and ChatGPT Work. Go build! And if you need inspiration, watch Astra in action, live: https://t.co/VmBLksCNZW
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Eric Wu, who built and ran Opendoor before stepping away in 2022, has had his new company, NavigateAI, out of stealth since May — building AI copilots that give construction workers real-time, hands-free guidance through smartphones and Meta's AI glasses, backed by $25 million from Elad Gil, Khosla Ventures, and Lennar to tackle a labor shortage severe enough that data center projects alone now need 4,000 to 5,000 workers apiece.
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Il round arriva pochi mesi dopo il pre-seed da 1,6 milioni di euro e sostiene la crescita della startup che applica l'IA alle gare d'appalto
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
MassRobotics has found that companies were split on the FCC ban on new foreign-produced mobile robots.
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Wang Naiyan has left to launch a physical AI venture, becoming the latest core tech talent from China's auto industry to enter the emerging field.
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Super interesting paper on proactive agents from Google DeepMind.
(bookmark it)
Proactive assistance usually means autocomplete.
Researchers asks what it looks like when an agent offers higher-level cognitive support and picks its own moment to speak.
They built a probe and deployed it with 16 …
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Join the discussion on this paper page
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Agility Robotics’ S-4 filing reveals $1.8 million in 2025 revenue, a $140 million operating loss and plans to scale its Digit humanoid robot.
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
A report on a blacklisted Chinese server maker's American subsidiar has added fresh uncertainty to a planned US-China AI summit set.
|
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
Universal multimodal embedding (UME) increasingly demands encoder's capacity for handling a broad range of tasks and modalities with increased complexity. Prior scaling methods either increase the representation size, retrieval effort, or scales the encoder into a heavy multimodal LLM. Recent works, such as Think-Then-Embed (TTE), explore scaling via reasoning tokens. However, embedding models are hard to scale up: increasing parameters directly tradeoffs for the large training batch size that contrastive learning needs, and retrieval has to be served under tight latency. Moreover, UME tasks are diverse in complexity, where scaling up embedders can bring significant redundant computation. In this work, we propose MOEMB, which instead scales UME along the expert axis through mixture-of-experts (MoE), growing encoder capacity while preserving single-vector, non-autoregressive encoding. Through a systematic study of the design space and training recipes for MoE-based UME, MoEMB sets a new state of the art on both MMEB-V2 and MRMR among models trained on public MMEB-family data: with only 3B active parameters, MoEMB surpasses TTE-based methods with >4x active parameters, using significantly less computes. To further improve the scalability and efficiency, we conduct the first comprehensive study of adaptive computation for MoE-based embedding, spanning diverse strategies across training-based and inference-only methods. Together, these results support expert scaling as an effective and efficient direction for UME, with adaptive computation further improving efficiency for MLLM-based embedding models towards large-scale retrieval and recommendation systems.
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
The viral "Ox Alpha" model is officially GLM-5.3-Flash from Zai. 🔍
📊 320B parameters with 18B active per token
⚡ Hybrid architecture mixing linear and sparse attention
💰 Leading performance on real world tasks from GDPval AA v2 at just $0.09 per task
The best part? Zai ran the massive free pre…
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
Meet Nex-N2.5 family, our latest open source agentic models.
Mini (35B) and Pro (397B) bring multimodal understanding and computer use. Max (1.6T) is a text-only MoE model built for complex reasoning, coding, and agent workflows.
Strong benchmark results in workflow automation and computer use:
…
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
The French AI lab has raised €3 billion at a €21 billion valuation in a Series D round led by Samsung, Scaleup Europe, and PSG Equity.
|
|
Scooped by
Philippe Gassmann
September 9, 5:02 AM
|
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the desc…
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Startup funding and venture capital opened the week with a strong tilt toward sovereign technology, physical infrastructure, vertical AI, and industrial automation. The standout deal is Indian space-tech company Pixxel’s $100 million Series C, the largest single funding round raised by an Indian space startup. In Australia, Jet Zero completed an A$30 million financing for
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Rudy Cohen, the co-founder and CEO of Inbolt, will reframe where physical AI pays off at RoboBusiness this fall.
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
One day in the near future you could be lying incapacitated in the middle of an earthquake-ravaged landscape where wheeled vehicles can’t reach you. Or in the very distant future, after engineers have solved thousands of problems associated with saving interplanetary astronauts from radiation, your…
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
// Normalized Low-Rank Adaptation (NoRA) //
They propose a one-line change to LoRA that costs nothing and improves convergence, stability and forgetting.
LoRA initializes the up-projection to zero, which means early optimization is governed almost entirely by the down-projection. This observation…
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
// Design Docs Are All You Need //
Banger paper from Google DeepMind, MIT, and colleagues.
What a genuinely strange and interesting paper this one is.
Here is the setup:
They maintain a performance-modeling library whose main branch contains almost no code.
The repository is a directed graph o…
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Join the discussion on this paper page
|
|
Scooped by
Philippe Gassmann
September 8, 3:04 AM
|
Broadcom beat Wall Street estimates on both revenue and profit. Shares fell anyway, and none of the coverage we found fully explains why.
|
![[2609.08634] Suan: Rectifying Direct Preference Safety Alignment in Large Language Models | AI News | Scoop.it](https://img.scoop.it/R3Nf-aV-zFCPDr3s1jV_-zl72eJkfbmt4t8yenImKBVvK0kTmF0xjctABnaLJIm9)
