safety

arXiv:2605.30576v1 Announce Type: new Abstract: Exploration in reinforcement learning for autonomous driving is inherently unsafe: agents must experience novel behaviors to learn, yet exploration can

arXiv:2506.22304v3 Announce Type: replace-cross Abstract: Continuous Normalizing Flows (CNFs) enable elegant generative modeling but remain bottlenecked by their iterative nature requiring costly samp

arXiv:2605.31521v1 Announce Type: new Abstract: Semantic speech tokenizers have become a widely used interface for Audio-LLMs, owing to their compact single-codebook design and strong linguistic align

arXiv:2605.31040v1 Announce Type: new Abstract: Developing effective representations for register transfer level (RTL) designs is crucial for accelerating the hardware design workflow. Existing approa

arXiv:2509.21561v2 Announce Type: replace Abstract: Ensuring the safety of surgical instruments requires reliable detection of visual defects. However, manual inspection is prone to error, and existin

arXiv:2605.31445v1 Announce Type: cross Abstract: In this work we study agents in simulated bargaining scenarios, where a buyer and a seller communicate through a text channel and attempt to negotiate

arXiv:2605.31131v1 Announce Type: cross Abstract: Attention-deficit/hyperactivity disorder (ADHD) is a psychiatric disorder which presents itself in individuals through patterns of developmentally ina

arXiv:2605.31524v1 Announce Type: new Abstract: Certification methods for stochastic systems provide sufficient proof rules, based on real-valued supermartingale certificates, to determine the almost-

arXiv:2605.30451v1 Announce Type: new Abstract: Group Relative Policy Optimization (GRPO) is an effective recipe for training reasoning models with verifier-based outcome rewards, but its supervision

arXiv:2605.31556v1 Announce Type: cross Abstract: Alignment teaches vision-language models (VLMs) to avoid expressing demographic biases, and when gender is clearly visible they largely succeed. Far l

arXiv:2605.30877v1 Announce Type: new Abstract: Large-scale Vision-Language-Action (VLA) pretraining is increasingly adopted as the foundation for robot policies, yet the evidence for pretrained VLAs

@webisticsdawg @GaryMarcus Personally I think everyone w a 401(k) should be absolutely livid. I'm disgusted by this, it's so brazen. How dare the richest man in the world pick working Americans' pocke

arXiv:2605.31561v1 Announce Type: new Abstract: Test-time reasoning has become a significant field of study since the introduction of chain-of-thought reasoning in large language models (LLMs). Howeve

What if the skeptics are right and superintelligence is impossible? Great! Then a proactive ban costs us nothing, prevents massive compute waste, and hurts no one. But if they are wrong? We face an un

arXiv:2605.30719v1 Announce Type: cross Abstract: We study when large language models (LLMs) can serve as effective black-box policy optimizers for reinforcement learning (RL) tasks, i.e., when can we

which is NOT new; see this quote from 5 years ago. the fact that all this is still true says a lot. This was right five years ago, and still is: “Large scale pretrained models are certainly likely to

arXiv:2506.00175v5 Announce Type: replace-cross Abstract: Modern AI systems are typically developed through multiple stages-pretraining, fine-tuning rounds, and subsequent adaptation or alignment, whe

arXiv:2605.31261v1 Announce Type: cross Abstract: The family of linear recurrent neural networks has shown strong performance as recurrent memory units in partially observable reinforcement learning.

arXiv:2604.01985v2 Announce Type: replace-cross Abstract: General-purpose world models promise scalable policy evaluation, optimization, and planning, yet achieving the required level of robustness re

arXiv:2603.10422v2 Announce Type: replace Abstract: World Models (WMs) offer a promising mechanism for post-training Vision-Language-Action (VLA) policies by providing dynamics priors that improve gen

arXiv:2605.30833v1 Announce Type: cross Abstract: On-policy distillation transfers reasoning capabilities by training a student model on its own generated trajectories using token-level feedback from

arXiv:2605.30612v1 Announce Type: cross Abstract: Continuous control policies trained with off-policy reinforcement learning frequently exhibit high-frequency action jitter, rendering direct deploymen

arXiv:2605.30896v1 Announce Type: new Abstract: Bidding in repeated auctions is a central challenge for reinforcement learning (RL), combining continuous control with the strategic complexities of dig

Anyone remember how I said in January 2025 that AI was going to be stumble and be dubbed “too big to fail”, along with cries for bailouts? If that call was correct – which increasingly seems likely, a

Gary Marcus discusses the Pope's understanding and perspective on artificial intelligence, examining statements or positions the religious leader has taken regarding AI technology and its implications

I am absolutely with @GaryMarcus and the Pope on this. (Not something you expect to say everyday). We are not creating beings. Systems don’t exp. grief nor hope, hold a value construct, or the ability

Listen, I’m a big “the index is the index” guy, but they are openly looting the coffers. This is 100% fraud. Rule changes for the SpaceX SPCX IPO: Index providers waived the profitability requirement

@ParValue26 @Hedgeye Let’s be clear what this actually is: the administration and the world’s richest man working together to screw over ordinary investors to goose returns on the IPO for the wealthie

serious accusation. does this fit with people’s experience? Is Anthropic altering model performance to force costly upgrades? Chapter Co-Founder and CEO @CobiBGantz outlines a shift his team recently

SpaceX being rammed into indices with no profit requirements, seasoning, and generally looser constraints is economic terrorism. Index trackers will eat the loss when reality catches up and retail inv

Gary Marcus argues that consumer backlash against AI-generated content will be increasingly visceral, particularly in creative fields like fiction, leading to growing demand for 'human-authored' certi

Three companies are about to IPO at a higher combined value than all 2,600 dot-com IPOs from 1995 to 2000 combined. – SpaceX, OpenAI, Anthropic (2026): ~3.75 trillion – Every dot-com IPO from 1995–200

Weird how the Pope seems to understand AI better than @geoffreyhinton, but I am 100% with the Pope on this. We are NOT creating beings. The Pope is right. We are creating interactive fiction that is t

weird the way this tweet was getting a ton of traffic and then just stopped. 🤷‍♂️ I honestly think Elon’s best days are behind him: BYD is crushing Tesla in EVs. Waymo is crushing Tesla in AVs. Anthro

AI safety can't happen behind closed doors! Super cool to see that the @AISecurityInst is releasing its evals, datasets, and models in the open on @huggingface, so researchers everywhere can scrutiniz

This post likely critiques the irony of someone insulting another person's intelligence while making a basic grammatical error (omitting an apostrophe in 'don't'). Gary Marcus, a cognitive scientist a

genius reply, full of reasoned intellectual argument, from the kind of retail investor who is probably going to get burned on the SpaceX IPO. Going to MIT and NYU may have meant something in the past,

Gary Marcus discusses his moderation approach on social media, noting that he blocks Tesla supporters primarily when they resort to insults rather than substantive arguments, and observes that this oc

I honestly think Elon’s best days are behind him: BYD is crushing Tesla in EVs. Waymo is crushing Tesla in AVs. Anthropic, Openai, and Google are crushing Xai on AI. The SpaceX S-1 is so ridiculous th

“I'm more concerned about the lack of intellectual diversity within the frontier AI commentariat/research world. This improved a lot over the last two years, but we're still far from a healthy ecosyst

is having a four month lead a sustainable multitrillion dollar business model? We took another look at the capability gap between open-weight and proprietary models. Since the start of the year, open-

it is the single most clever move Elon ever pulled off Rule changes for the SpaceX SPCX IPO: Index providers waived the profitability requirement and cut the seasoning window from 90 days to 5. This f

More of my thoughts on the topic in @Corriere: https://www.corriere.it/esteri/26_maggio_29/yoshua-bengio-ai-pope-6aa95de2-33f7-4f73-8905-34b1abedbxlk.shtml “Like nuclear energy, AI must be at the serv

Gary Marcus expresses frustration about an influx of Elon Musk supporters on his social media feed who conflate personal insults with substantive counterarguments. The post appears to critique a lack

so much for multitrillion dollar candy companies Note to all staff: Turns out that super tasty AI candy we've been putting out in large bowls isn't free and it isn't necessarily leading to better prod

this won’t end well. it will end with a bailout. Cash flow no longer covers the AI capex bill, so hyperscalers are funding it with record debt: Hyperscaler bond issuance has soared to $150 billion YTD