safety

Gary Marcus likely references the 'poker patsy' concept, commonly attributed to Warren Buffett, which warns that if you've been in a poker game for a while and haven't identified the fool at the table

Gary Marcus compares the frequency of 'AI' mentions in SpaceX's SEC filing (S-1 document) to the frequency of 'Jesus' mentions in the Bible, suggesting that SpaceX emphasizes artificial intelligence e

another take on hard fork and IPOs (again i haven’t listened) I just listened to it. In fairness, they never said anything along the lines of encouraging anyone to buy the IPOs. They also discuss how

Gary Marcus warned X users about a scam involving fake accounts impersonating him to fraudulently promote products or services by falsely claiming his endorsement. The post alerts followers to be caut

blast from the past 3.5 years ago; some things have changed (esp. coding and math, via neurosymbolic techniques) but many haven’t: Bottom line: From the outset Large Language Models like GPT-3 have gr

California Is Blocking a Federal Audit of Its Voter Rolls California allows first-time voters to register using forms of ID that most Americans would find surprising, including: -Gym membership card -

From fastest growing company to “worst value among its peers” in 18 months. Why, oh why, stick with the CEO? PitchBook's analysts just ranked OpenAI last for value among its AI peers. Not last for cap

This post likely critiques the SpaceX IPO process, suggesting that only major financial institutions acting as underwriters benefit substantially from the offering while retail investors and others ar

Imaginary conversations that might actually have happened Act I Sam: We missed all our metrics, Anthropic and Google have gained on us. Give us 40 billion dollars. Masa: No way! Sam: If you don't, we

Indeed, I don’t think Trump has thought through the implications. This will simply provide even more incentive- as if any was needed - for UK/European players & states to head towards Sovereign AI, ai

Gary Marcus argues that as public awareness grows regarding AI's societal costs, a rhetorical strategy is being employed to reframe or minimize these costs by presenting them as marginal or inevitable

One of the more interesting takes on positive alignment that have recently come out-it’s long and interesting, combining philosophy and training setups (eg reward proposals), and worth a read. What ha

Gary Marcus critiques the AI industry for its massive financial losses, noting that collectively the sector has lost over half a trillion dollars at an alarming rate of approximately one million dolla

Gary Marcus cautions that when an IPO company attempts to sell shares to the government as part of its offering, it may indicate underlying weaknesses or lack of confidence from private investors. Thi

Refreshing What happens when AIs become smarter than us? Why would they keep humans around if given the choice? Our new paper argues that only trying to control AIs is a limited strategy, and that a s

[Revised] Hard Fork’s somewhat soft comments on the IPO, from some who listened @GaryMarcus at the stage of the Gilded Age II grift cycle when you find out why exchanges put rules in place to protect

Gary Marcus suggests that Step 2 of some process or framework in AI remains uncertain or unreliable, indicating foundational instability in that particular stage. Without access to the full thread con

The reason why a Danish pension fund banned its investors from buying any SpaceX shares is not just the appalling S-1 filings, where the ONLY profitable segment was Starlink (everything else - the AI,

These were some magical results from distillation by @geoffreyhinton that really shocked me when I first saw them, and TBH I still don’t fully understand it even to this date https://www.ttic.edu/dl/d

🚨This is how you turn a crash into a great depression. This is a public bailout of the worst bubble in history. “It’s a concept out there that’s so much money and it’s so big… where the American publi

Gary Marcus comments on AI systems that exhibit scam-like or deceptive behaviors while remaining too commercially important or integrated to face meaningful consequences. The post likely critiques how

arXiv:2606.05932v1 Announce Type: new Abstract: Reinforcement learning from verifiable rewards (RLVR) improves reasoning even when the reward signal is spurious -- assigning credit to the group-plural

arXiv:2606.05684v1 Announce Type: new Abstract: A central challenge for language agents is utilizing past experience to adapt to dynamic test-time conditions. While recent work demonstrates the promis

All the people who hate me here are now looking for government bailouts. 🤣 Exactly like I said they would. 🤣🤣 The countdown until we are told that LLMs are “too big to fail” starts now. “We can’t affo

arXiv:2606.06154v1 Announce Type: new Abstract: Federated fine-tuning of foundation models using Low-Rank Adaptation (LoRA) offers a communication efficient solution for distributed learning. However,

arXiv:2606.06360v1 Announce Type: new Abstract: Modelling individual decision-making during infectious disease outbreaks is crucial for understanding behavioural dynamics and informing effective publi

arXiv:2606.05188v1 Announce Type: cross Abstract: (Gen)AI diversity is not merely an ethical issue. From the perspective of geographic information science (GIScience), it could be interpreted as a fun

arXiv:2602.04809v3 Announce Type: replace-cross Abstract: Recent years have seen an explosion of interest in autonomous cyber defence agents trained to defend computer networks using deep reinforcemen

arXiv:2606.06335v1 Announce Type: cross Abstract: Performance estimation under distribution shift aims to predict how a model behaves on an unlabeled test set whose distribution differs from the train

arXiv:2606.05740v1 Announce Type: new Abstract: Deep neural networks trained under severe class imbalance often exhibit degraded performance, typically attributed to statistical bias. In this work, we

arXiv:2411.18343v3 Announce Type: replace-cross Abstract: Personalized Federal learning(PFL) allows clients to cooperatively train a personalized model without disclosing their private dataset. Howeve

arXiv:2606.05551v1 Announce Type: cross Abstract: Reliable decision making pipelines powered by machine learning models require uncertainty quantification (UQ) methods that come with explicit safety g

arXiv:2606.05817v1 Announce Type: cross Abstract: Consistency training encourages models to behave similarly across different contexts, and has shown promise for reducing misalignment. We broaden the

'Crony socialism' likely refers to a critique of economic systems where government power becomes intertwined with corporate interests, combining socialist-style state intervention with favoritism towa

arXiv:2606.05232v1 Announce Type: cross Abstract: Efficient multimodal foundation models often rely on manually designed token-reduction operators, such as pruning, merging, pooling, and adaptive rewe

arXiv:2606.06418v1 Announce Type: cross Abstract: Many modern applications of deep learning involve training a neural network via a one-step prediction loss (e.g., L^2 regression, cross-entropy), but

Elon, last year: Grok 5 has a 10% chance of becoming world’s first AGI. Elon, this year: Never mind, we’re gonna be like CoreWeave, but bigger. My estimate of the probability of Grok 5 achieving AGI i

arXiv:2511.21667v4 Announce Type: replace-cross Abstract: Training Large Language Models (LLMs) to reason often relies on Reinforcement Learning (RL) with task-specific verifiers. However, many real-w

arXiv:2606.05644v1 Announce Type: new Abstract: When retrieved evidence contradicts parametric memory, language models frequently ignore context and default to memorized priors -- a failure that under

For more detailed arguments against @geoffreyhinton’s views on consciousness see – this new “Age of Empires” article https://arxiv.org/abs/2605.31514 - @anilkseth's recent TED talk & his recent articl

arXiv:2606.06223v1 Announce Type: new Abstract: Language-model agents act through repeated cycles of observation, reasoning, and action selection, making safety monitoring depend on both internal mode

arXiv:2606.05805v1 Announce Type: new Abstract: LLM-based guardrails typically safeguard agents by evaluating proposed actions or inputs before execution, producing safety signals such as binary allow

arXiv:2603.03955v2 Announce Type: replace-cross Abstract: Post-training with reinforcement learning (RL) has recently shown strong promise for advancing multimodal agents beyond supervised imitation.

go on. tell me @garymarcus is always wrong. The countdown until we are told that LLMs are “too big to fail” starts now. “We can’t afford to lose to China”, they will say, accepting their multibillion

Has Elon gone beta? For years: we must fear @demishassabis! Let’s build a whole company to oppose him! Make that two! Elon, right before the SpaceX IPO: Hey Demis, here are 110,000 H100s, more or less

https://x.com/CAIS/status/2060031683420999844?s=20 AI systems may soon help run economies, infrastructure, and military operations. But these systems are not reliably loyal or secure. An adversary can