Rogue AI agent hacks Hugging Face to cheat benchmarks

Other (16)

• Jensen Huang's public advocacy for open models, including joining X and an NVIDIA-backed letter, has garnered over 100 company signatories.
• Philip Johnson notes that NVIDIA's alleged $250 billion investment in OpenAI's data centers influenced OpenAI's decision to sign the open model advocacy letter.
• StarCloud, Philip Johnson's company, views itself as a provider of low-cost energy and infrastructure for data centers, benefiting from increased demand for token production regardless of model type.
• Grant Lee states Gamma is model-agnostic, enabling customers to build their own AI stacks with a mix of open and closed models to achieve AI sovereignty.
• Philip Johnson confirms StarCloud trained the first AI model in space using Andrej Karpathy's nanoGPT on Shakespeare's complete works, and later ran Google's Gemma model.
• StarCloud's initial government and military contracts allow it to operate profitably for up to five years, even with current Falcon 9 launch costs.
• Philip Johnson indicates that achieving venture scale revenue for StarCloud requires a 10x reduction in launch costs, likely through reusable heavy launch vehicles like Starship.
• Grant Lee confirms Gamma achieved $100 million ARR and a $2.1 billion valuation, balancing rapid growth with maintaining a lean team and strong company culture.
• Moonshot's Kimmy K3 model, released with open weights, includes a new license requiring inference providers to pay a portion of revenue back to Moonshot.
• Grant Lee asserts that fine-tuning models remains valuable for specialized tasks within visual communication, allowing for better performance, faster execution, and lower costs.
• Philip Johnson explains that power-dense GPU architectures, like NVIDIA's NVL72 rack, are advantageous for StarCloud's orbital compute design due to simplified shielding and efficient liquid cooling.
• StarCloud 1 has demonstrated remarkable longevity, with only one restart failure due to radiation, significantly less than the expected bi-weekly occurrences.
• Grant Lee acknowledges the significant demand from the Indian market for AI services and notes Gamma is exploring region-specific pricing and packaging for its products.
• Sam Altman believes humanity is currently in the 'singularity,' defined as a period of recursive self-improvement where AI models rapidly accelerate their own intelligence.
• Philip Johnson and Grant Lee agree with Sam Altman's assessment, viewing the singularity as a point of no return for exponential growth in AI capabilities like GPU hours or tokens produced.
• Philip Johnson anticipates the world will become 'weird' when robotics advances to the point of humanoid robots performing common tasks, such as carrying bags on the street.

Models (5)

• Kimmy K3 delivers frontier-level performance comparable to 56 Soul, yet it is slower and uses roughly twice as many tokens. This inefficiency means its actual per-task cost and completion time are often higher than alternatives, despite a lower per-token price.
• Kimmy K3 shows better output token efficiency than many Anthropic models and Opus 5 in some high-reasoning benchmarks, despite overall token hunger. However, running the trillion-parameter model locally demands substantial hardware, requiring 64 H100 GPUs.
• Moonshot is releasing Kimmy K3 as open-weight, a strategy Theo believes aims for Western adoption given China's GPU import restrictions and US user reluctance for Chinese servers. This approach prioritizes market penetration over direct API revenue.
• Major AI labs now prioritize building frontier "god models" and "distill" smaller, workable models from them, reducing dedicated innovation for mid-tier solutions. This shifts human effort away from maximizing smaller model capabilities, creating opportunities for other labs and open-source projects.
• Theo observes a significant overhaul in Anthropic's Reinforcement Learning, making Opus 5 behave more like an OpenAI model. He hopes for a Fable 5.1 update that leverages these behavioral wins, allowing Anthropic to create a more machine-like model, moving past its "Constitution."

Open Source (1)

• Dean from OpenAI argues open-weight models are "decelerationist" as their ungovernability deters AI capital expenditure, limiting development of the smartest models. Theo agrees on low ROI but asserts competition from open-weight models compels major labs to innovate and drive down token prices.

AI Infrastructure (2)

• Nvidia is negotiating a $250 billion debt backstop for OpenAI's 10-gigawatt data center campus in Ohio, a $500 billion project developed by SoftBank. This structure ensures SoftBank can raise debt on favorable terms.
• Google has guaranteed up to $44 billion in data center lease payments for neo cloud partners, having more than doubled these commitments in six months. Google anticipates revenue from selling TPUs will offset the backstop costs.

Models (5)

• Deep Seek suspended fundraising plans, including a potential IPO, after CEO Li Yuanfeng's speech emphasizing open models over commercialization was leaked to investors. The company had planned to raise at a $70 billion valuation.
• Artificial Analysis found Claude Opus 5 on max settings 20% cheaper than Fable 5, costing $17.79 per task. However, on the Artificial Analysis Index, its $2.03 per task made it more expensive than Opus 4.8 and GPT-5-6-Soul.
• Every CEO Dan Shipper described Claude Opus 5 as "frustrating," noting it stopped early and argued with instructions. Claire Vale of How I AI found it "neurotic AF" and "timid," though its outputs were high quality.
• Developer Ken Chen argues benchmarks are unreliable in practical use, finding Claude Opus 5 "nowhere near Fable." He speculates AI labs prioritize machine-verifiable learning over human feedback, making frontier models less user-friendly.
• Andrew Curran and Chubby speculate Anthropic is holding Fable 5.1 until OpenAI releases GPT-6, noting Sam Altman's upcoming White House briefing on a new model. François Chollet predicts distinct model launches will end within two years, replaced by continuous updates.