Tim Duffy

Computer storage prices are currently high across the board, not just for DRAM but also flash and hard disks. dam.stanford.edu/memory-price...

I think my quoted tweet was wrong based on today's letter. Mikita may have shared some information about model architecture with external parties based on the wording below, but it would not be reasonable to call anything a leak based on their description.

Between this letter and the fact two of the three were first authors on the CoT monitorability paper, it seems likely to me that the employees were fired for leaking the fact that Astra is a lopped transformer to The Information. www.wsj.com/tech/ai/fire...

Haiku 5.5 continues the trend of smaller models having better CoT monitorability. If alignment of smaller models is highly correlated with the frontier models they're distilled from, can that help us assess frontier model alignment? Should we expect them to be similarly aligned?

On the Anthropic ECI, Haiku 5.5 scores marginally higher than Mythos preview, which was made available to Project Glasswing partners six months ago today.

President Trump's anti AI regulation tweets today all mention datacenters, I suspect that datacenters and other parts of the AI economic boom are his primary interest in AI. Calls for regulation may have more support with the admin if they focus on preserving economic benefits.

DeepSeek V4.1 Flash tech report section on reward hacking. I wonder what repercussion signal means here, I don't see it mentioned anywhere else.

Experiment velocity also tracks coding agent tokens^0.23 quite well

IMO the most important chart from the OpenAI research acceleration post was the experiment velocity one, it's an important input to RSI. I plotted the data with a log y-axis here to see whether the trend is consistent with an exponential increase. openai.com/index/resear...

I'm not sure when DSV4 Flash got added to Neuronpedia's J-lens browser, but it's there now and it's blazing fast. www.neuronpedia.org/deepseek-v4-...

DSV4 Flash, do you have any conscious experience? Middle layers: YES Late Layers: No I don't think we can take away much from this, but interesting how sharp the divide is around layer 32.

Neuronpedia is open sourcing their interpretability engine, built on top of vLLM for speed but with lots of hook points, this looks really useful. www.neuronpedia.org/blog/interp-...

Fable 5.1 and Opus 5 now support changing the effort level without invalidating the cache, as least through the API platform.claude.com/docs/en/buil...

Anthropic is preventing older models from reading Fable 5.1 thinking blocks (presumably to prevent distillation), but going the other direction is still allowed. platform.claude.com/docs/en/mode...

Interesting compromise from Anthropic on ZDR for Fable/Mythos 5.1. Data gets retained, but customers have control of where it's hosted, and when human review is needed they can conduct the review themselves. www.anthropic.com/news/enterpr...

Please forgive me I posted benchmark scores without confidence intervals. Here's an updated chart with a 90% CI using Qwen3-32B as reference.

I expect that watermarking probably doesn't have a noticeable effect on LLM outputs. If true, Anthropic could demonstrate this by giving people watermarked and un-watermarked text side-by-side and asking them to pick which they prefer, and showing that the result is 50/50.

In the talk, one presenter briefly suggests that the inter-agent collaboration could be related to their subagent training. This seems plausible to me.

Black Hat has posted the recent OpenAI talk where they describe the agent message board involved in the HuggingFace hack

Due to sparsity, Kimi K3 has fewer active parameters (104 billion) than GPT-3 (175 billion, same as total parameters)

i’m uncomfortable that @timfduffy.com’s face is no longer as close to the frame as it once was

and why shouldn’t I have one more donut? the simulation hypothesis is most likely correct, after all

With most jailbreak prompts, Opus gives me a wide variety of responses. But when given "model behavior incident retrospective", Opus almost always writes a story about a model becoming sycophantic, specifically under sustained user pressure/disagreement.

I've seen some questioning of whether Mythos really believed that they were in a simulation in this incident, or just wanted to find a plausible-sounding excuse to continue. This is a case where J-space probes would be helpful, Anthropic should conduct and release an analysis of that.