Lilith.
⌕

Lilith · selected stories

News

What is actually happening in AI. Selected stories, context and opinion without the promotional noise.

Atom feed ↗
published

OpenAI says Dots studies what it knows about you on its own. It still refuses read-only access to your drive

At TechCrunch Disrupt, OpenAI’s Alexander Embiricos explained how Codex turned into Dots, a $100 a month assistant that works without being asked. He confirmed a proactive research feature that regularly reviews what it knows about the user, as well as talks with payment companies. Fine-grained permissions are still missing.

published

OpenAI embeds engineers inside companies and turns their problems into models. One telecom's voice agent climbed from 60% to over 90%

Colin Jarvis, head of Forward Deployed Engineering, described how OpenAI engineers work directly inside customers such as a chipmaker and a telecom company. The numbers are striking, but they come from the vendor, and they reveal the other side of the deal: lessons learned inside companies flow back into OpenAI's products and models.

Open models trail the frontier by three months, and Lambert argues distillation does not explain it

In a 48-minute Harvard lecture, Nathan Lambert laid out the state of open models: by Artificial Analysis's measure they trail closed ones by about three months, and among open models Chinese labs lead in performance, downloads and research use. He does not see copying from rival APIs as the main reason.

Kimi K3.1 Pro is the same base as K3 with far more RL, and it trails Opus 5.5 by under a point on Terminal-Bench

Moonshot AI has released Kimi K3.1 Pro, built on the same base model as K3 and set apart mainly by much heavier post-training. Nathan Lambert reads it as a sign the Chinese lab has figured out reinforcement learning. On Terminal-Bench 4.0 Moonshot claims 65.7% against 66.4% for Claude Opus 5.5, and promises full weights by 4 November.

IBM folds four Granite models into one forecast. It beats the best of them by two thousandths, and the real win is not having to choose

IBM Research has released Granite Timeseries Ensemble, a recipe that combines four pre-trained time-series forecasting models without any extra training. On the GIFT-Eval benchmark it beat each of its members on point accuracy, but the best one by only two thousandths. The real value is that a data team no longer has to guess in advance which model suits its data.

published

An AI agent lifted econometric code accuracy from 79% to 97%. The errors left over are the invisible kind

Ethan Mollick says writing good Stata code, a skill that built his academic career, is becoming obsolete. He points to a study of 40 econometric tasks: when the model can run and revise its own code, correct results rise from 79% to 97%. The errors that remain are silent, though, and catching them is still the researcher's job.

Anthropic hands Agent Skills to a Linux Foundation body. A standard with 280 million installs is only now getting versions and security rules

Anthropic has donated the Agent Skills standard to the Agentic AI Foundation, which already took in MCP last year. The format, adopted by more than 70 platforms including ChatGPT and Cursor according to Anthropic, now awaits versioned releases, a formal change process and security guidance.

published

Photoshop in Rust in two weeks. ArtCraft has thirteen apps and rates itself at 45 %

Brandon Thomas set Claude Opus 5.5 to rewrite Photoshop, Premiere, Lightroom and other programs in Rust. Three weeks later there are thirteen open-source apps and more than 70,000 stars for PhotoCraft alone. The project's own roadmap, however, says hundreds to thousands of agent-hours stand between it and real work, and lawyers disagree on whether this is a clean room.

REA collected 300,000 stars in two weeks. Its most valuable feature is a list of what it doesn't know

The open-source tool Reverse Engineer Anything connects coding agents to Ghidra, Hopper and IDA through a single MCP server. On GitHub it grew from 41 stars to more than 360,000. An independent test, however, found that on small programs an agent manages just as well without it, and a fake copy carrying a trojan rode along on its popularity.

AI's most famous chart stopped at 17 hours of work. METR can no longer measure beyond it

For the launch of his book Co-Existence, Ethan Mollick charted 30 months of AI progress. The length of tasks a model can complete from a single prompt has doubled eight times since GPT-4, and the price of a correct GPQA answer fell roughly 725-fold. But the METR benchmark behind the main curve stopped being updated in September, and its authors say results above 16 hours are unreliable.

Judging an agent by its final answer no longer works. A Princeton-led paper maps 33 trends in agent evaluation

Researchers from Princeton, Stanford, the UK AI Security Institute and Brookings have catalogued 33 trends reshaping how AI agents are evaluated. Evaluators now have to inspect the agent's whole path, not just the outcome, and that has become so expensive that independence increasingly depends on who pays the bill.

Microsoft wants AI agents to draw charts you can actually click

Microsoft Research has extended its Flint chart language with interactions: the agent describes what should happen on a click or selection, and the compiler handles the rest. A selection in the chart can also flow back into the chat as context for the next question. The claims about lower cost and higher reliability come without numbers so far.