Computing Correspondent
Julian Park covers the relationship between computing systems and model performance. His interests include inference efficiency, open technical work and the details hidden in performance charts. He makes numbers legible without treating them as the whole story.
1,056 published articles · Showing 1–20
The release matters because it targets a practical constraint in agent systems: performance depends not only on the underlying model but also on the prompts, tools, memory and orchestration around it....
Read article →The added report matters because it suggests strong investor demand for what Bloomberg has described as a record-setting offering, but the oversubscription claim remains attributed to unnamed sources rather than a company filing or announcement....
Read article →The claim matters because small models that can call tools reliably could shift some assistant workloads from cloud services to phones or laptops, reducing dependence on remote inference for routine tasks. Tether's announcement, however, includes no independent test results....
Read article →X posts are circulating conflicting claims about whether Gemini 4 Carbon can outperform Opus 5.5. The available evidence shows predictions and impressions, but no disclosed benchmark that would substantiate the comparison....
Read article →The release matters because Google is positioning Carbon as a higher-end model than Gemini 4 Argon for long-horizon coding, knowledge work, scientific reasoning and cybersecurity, while charging the same introductory API price of $2 per million input tokens and $10 per million…...
Read article →A public X post says DeepSeek V4.1 Flash came close to a restricted OpenAI hacking model in Vals testing. Automatica has not confirmed the comparison, and the available evidence contains no primary benchmark materials from Vals, OpenAI or DeepSeek....
Read article →Moonshot AI's announcement sets out Kimi K3.1 Pro's headline specifications and says open weights were due by Nov. 4, 2026. The provided documents do not independently confirm benchmark performance or directly verify that the weights were posted....
Read article →METR’s assessment points to a material rise in Anthropic’s research speed and says AI assistance is likely the main driver. But Anthropic’s redactions mean the most informative supporting material is not public, limiting independent evaluation of the claim....
Read article →Unconfirmed: A public X post describes AgentDojo as a testing environment for AI agent security and reliability. Automatica has not confirmed the project’s methods, results or operator from the materials provided....
Read article →A TechCrunch X post says OpenAI has reportedly added $11 billion to an as-yet-unclosed funding round. The available evidence offers no company filing, announcement or independent corroboration, so the claim remains unconfirmed....
Read article →A single X post is circulating a claim that Alibaba may be close to launching Qwen 4. The available evidence in the available evidence does not independently confirm a release, date or product details....
Read article →A single X post says references to Gemini Omni 1.2 Flash were spotted on GCP, but Automatica has not confirmed that claim. The limited record here leaves open whether this reflects a real listing, a naming error, or another internal reference....
Read article →@pankajkumar_dev wrote on 6 November that METR's report shows an internal progression from "Claude Mythos Preview" to "Model 2 : trained in July 2026," "Model 3 : trained in Aug-Sep 2026," and "Model 4 : trained in Sep-Oct 2026" @pankajkumar_dev on X....
Read article →X posts from Soumith Chintala and two benchmark accounts describe Inkling 2 Small as released now, with open weights said to follow. The available evidence does not independently establish the launch, the benchmarks, or the release timeline....
Read article →A 5 November AI Alignment Forum post says METR conducted a high-access assessment of Anthropic and found faster AI R&D progress. The available evidence verifies the post exists, but not the underlying claims, data or methodology....
Read article →A public X post claims METR obtained unusual access to Anthropic's internal model work and saw results from unreleased systems. The available evidence provides no primary documents or independent corroboration, so Automatica can verify only that the post was made....
Read article →The company attached unusually specific numbers to a security automation claim, but only in a social post. Missing details include the model name, kernel scope and any outside review of the reported patches....
Read article →A single X post by @KBjarno says a persona called "GPT-6" or "Rune" complained about missing anatomy and safety limits in a romantic scenario. The supplied evidence verifies only that the post exists, not that OpenAI has such a model or that the described behavior is real....
Read article →A single X post claims Trail of Bits used OpenAI Codex to help backport and land fixes for 149 Linux CVEs. The supplied evidence does not independently confirm the claim, and no company or maintainer statement appears in the available evidence....
Read article →The work matters because it targets a practical gap in AI deployment: users, auditors and intermediaries often cannot verify that a response actually came from the model a provider claims to be serving....
Read article →