Script
Here is today's Hacker News Daily for Tuesday October 6th. Reflection has announced Beam, a 501-billion-parameter open-weight mixture-of-experts model, with 23 billion parameters active per token. The company is positioning it for coding, reasoning, and agent-style workflows, saying it is competitive with leading open models while requiring less inference compute. The weights, technical report, and model card are promised later this month, after final evaluations and red-teaming. [1]
HN welcomed another serious open-weight entrant, but much of the thread was about what remains unknown. Commenters questioned the vague reference to proprietary licensed training data, the presentation of benchmark charts, and what hardware Beam will actually need in practice. One criticism was that the charts visually buried stronger competing open models. Others argued that Western labs are still chasing Chinese releases that are smaller, cheaper, and already available. The vibe was interested but unconvinced: people want reproducible evaluations, weights, and deployment details—not just giant training-run numbers. [2]
A Nieman Lab report highlighted a more immediate AI problem: ChatGPT-generated cartoons have been appearing with the signatures of real New Yorker cartoonists. Brendan Loper, who signs his work “BLOPER,” was contacted by people who assumed an AI-generated Dolly Parton cartoon was his. Other cartoonists have found similar false signatures attached to work they neither made nor endorse. [3]
The thread largely agreed this is a bug worth fixing, but disagreed on the right framing. Some called it copyright washing; others said false attribution and trademark-like impersonation are the more relevant issues. A practical question was whether generated cartoons should bear an AI label, a model signature, or no signature at all. One artist compared it to having “somebody attributed a quote to me that I didn’t say.” The overall mood was unusually clear-eyed: even people who accept generative tools saw a forged-looking artist signature as a harmful and avoidable failure mode.
Another popular post claimed that Opus 5.5-based agents identified two candidate room-temperature magnetic semiconductors through literature research and quantum-mechanical simulations. One candidate was newly designed; the other was first synthesized in 1999. If experimentally confirmed, such materials could matter for spintronics and future memory technologies. [4]
But HN was careful not to call this a discovery yet. Commenters stressed that the agents appear to have orchestrated standard density-functional-theory simulations, rather than inventing a new physical method, and that predictions are only the beginning. “Until actually made and tested, not worth getting excited over,” was the prevailing view. Some readers were enthusiastic about AI making materials exploration cheaper and faster; others recalled the LK-99 episode and urged extreme caution. The vibe was technically curious, but strongly allergic to hype.
Finally, an essay on Apple, macOS security, and autonomous agents sparked a broader argument about who should control a computer. The author’s Mac was compromised through an actively exploited Screen Sharing vulnerability after port 5900 was exposed to the internet. The essay also examined Apple’s tighter treatment of Full Disk Access, especially as agent software asks for broad permissions. [5]
Commenters split between seeing Apple’s restrictions as sensible protection against prompt injection and invasive software, and seeing them as another move toward platform control. The strongest consensus was simpler: agents that can read messages, files, and browsing history need permissions users can understand and limit. Across these threads, HN’s recurring demand was accountability—whether for model claims, AI-made media, scientific predictions, or software acting on a user’s behalf. Thank you for listening to Hacker News Daily from The Daily FM. See you tomorrow!
- Beam: Reflection's 501B open-weight model
...to the model. We will release the weights, technical report, model card, and developer artifacts later this month.Model CapabilityWe trained Beam with a particular focus on coding and agentic performance. Beam advances the Western open-weight frontier and is competitive with larger open models like GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks. Where frontier open models like Kimi K3 remain ahead on raw capability, Beam's advantage is efficiency at inference time.The below figure shows Beam's performance across a Discussion — top comments: htrp: > Beam is a sparse Mixture-of-Experts model with 501 billion total parameters, 23 billion active, built for coding, reasoning, and agentic workloads. > Beam’s capabilities come from major investments in both pretraining and reinforcement learning (
- Beam: Reflection's 501B open-weight model
...pen base models. In parallel, we developed the algorithms, training environments, and infrastructure needed to sustain high-compute RL at exceptional scale. Our high-compute RL run generated over 100 million rollouts on 10.5K NVIDIA GB300 GPUs over 4 weeks of training.Together, these efforts produced competitive open-weight performance with frontier inference compute efficiency.Beam is undergoing final red-teaming and evaluations. You can sign up here for early access to the model. We will release the weights, technical report, model card, and developer artifacts later this month.Model CapabilityWe trained Beam with a particular focus on coding and agentic performance. Beam advances the Western open-weight frontier and is competitive with larger open models like GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks. Where frontier open models like Kimi K3 r...
- ChatGPT is adding real cartoonists' signatures to fake New Yorker cartoons
...omments section of her post, she wrote she had simply asked ChatGPT to make “a New Yorker-style cartoon.” “I had people writing to me about this cartoon, asking if it was mine,” Loper told me. First came a text from his twin brother, whose friend had seen the AI-generated image. Emails and DMs on social media from strangers followed. “It was very surreal,” he said. This wasn’t the first time ChatGPT had reproduced Loper’s signature. In May, a stranger reached out over Instagram to warn Loper that his pen name had appeared in an image output by the chatbot. Over the summer, Loper found additional examples in a Reddit forum about ChatGPT, where users had been posting similar New Yorker-style cartoons they had generated. Loper already knew his work was “ Discussion — top comments: dormento: The copyright washing machine strikes again. > drdaeman replies: That’s not a c...
- Opus 5.5 agents discover two room-temperature magnetic semiconductor candidates
...hat these extremes mean. Today, I’ll share what we found. A team of AI agents and I designed one candidate magnet and found another, first made in 1999, that our calculations predict has the properties we were after. But before diving into the details, let me first lay out a magnet primer that takes all of 90 seconds, assuming you are not an undergraduate in physics or chemistry. A 90 second primer in magnets Spin Each electron has a quantum mechanical property called ‘spin’, which is responsible for its magnetic moment. We can model the spin direction for each electron as either pointing up or down. Spintronics We use spin for storage. A magnetised material stores information based on the spin up/spin down orientation of its electrons, in the same way classical magnets store information based on pointing up/down. The most prominent example of spintronics is the hard...
- Apple and a hacker's future
...rned that a high-severity macOS vulnerability that allows attackers to execute malicious code is under active exploitation. “The NCSC has received a notification indicating that active abuse of this vulnerability has been observed on multiple systems on which port 5900 was accessible from the Internet,” the Netherlands National Cyber Security Centrum warned earlier this week. “In all these cases, root had been accessed on the affected system and a Monero crypto miner had been placed.” The vulnerability, tracked as CVE-2026-65400, received a patch from Apple last week for macOS Tahoe, Sequoia, and Sonoma. The vulnerability, with a severity rating of 7.1 out of 10, stems from a bug in the macOS screen sharing capability, which allows a remote party to view the screen and control the keyboard and mouse while a machine is turned on. A flaw in the “state management,” whic...