Chris Clark

When I’m king, gate numbers will always correspond to how far away they are from the terminal entrance

This meeting is a waste of my tokens

people are walking around with their laptops slightly ajar to keep their agents running

We built a great AI writing detector but unfortunately it’s not very scalable. @pingToven can only read so much :(

Moonshot’s Kimi K2.6 is the new leading open weights model. Kimi K2.6 lands at #4 on the Artificial Analysis Intelligence Index (54) behind only Anthropic, Google, and OpenAI (all 57) Key takeaways: ➤ Increase in performance on agentic tasks: @Kimi_Moonshot's Kimi K2.6 achieves an Elo of 1520 on our GDPval-AA evaluation, which is a marked improvement over Kimi K2.5’s Elo of 1309. GDPval-AA is our leading metric for general agentic performance, measuring the performance on knowledge work tasks such as preparing presentations and analysis. Models are given code execution and web browsing tools in an agentic loop via our open source reference agentic harness called Stirrup. This continues Kimi K2.6’s strength in tool use, maintaining a 96% score on τ²-Bench Telecom, placing it among other frontier models in this category. ➤ Low hallucination rate: Kimi K2.5 scores 6 on the AA-Omniscience Index, our knowledge evaluation measuring both accuracy and hallucination rate. This score is primarily driven by a comparatively low hallucination rate of 39% (reduced from Kimi K2.5’s 65%), indicating a greater capability to abstain rather than fabricate knowledge when the model is uncertain. Kimi K2.6’s low hallucination rate places it similarly to other models such as Claude Opus 4.7 (36%) and MiniMax-M2.7 (34%) ➤ High token usage: Kimi K2.6 demonstrates high token usage, but is in line with other frontier models in the same intelligence tier. To run the full Artificial Analysis Intelligence Index, Kimi K2.6 used ~160M reasoning tokens. This is slightly lower than Claude Sonnet 4.6 (~190M reasoning tokens) but much higher than GPT 5.4 (~110M reasoning tokens). ➤ Open weights: Kimi K2.6 is a Mixture-of-Experts (MoE) model with 1T total parameters and 32B active, same as the previous two generations of models Kimi K2 Thinking and Kimi K2.5. Kimi K2.6 again pushes the open weights frontier in intelligence. ➤ Third Party Access: Kimi K2.6 is accessible through Moonshot’s First Party API as well as third party API providers Novita, Baseten, Fireworks, and Parasail ➤ Multimodality: Kimi K2.6 supports Image and Video input and text output natively. The model’s max context length remains 256k. Further analysis in the threads below.

See ya later calculator. In a while, numberphile.

@aviel Is this like an elaborate way of saying that I can see right through your bullshit?

I’m spread so thin that you can see through me. No room for bullshit.

Seems like when most people talk about ARR they really mean ARRR - annualized revenue run-rate. I propose we start using this metric more broadly, and that we distinguish between it and ARR by saying ARRR in a pirate voice.

The commercial relationship between the labs and clouds is also apples and oranges. The OpenAI/Azure relationship is an IP licensing agreement with a rev share, whereas the Anthropic/Hyperscaler relationships look nothing like that. Therefore they are not accounted for the same way.

The way @OpenAI and @AnthropicAI account for revenue / ARR is apples to oranges. Should Anthropic treat their revenue from AWS and other hyperscalers the same as OAI, they would be a materially lower in rev… If they both IPO in the coming quarters, not sure how the SEC is going to let these two companies have different accounting treatment for essentially the same type of revenue. OpenAI TAKES OUT the 80% revenue share that goes to @Microsoft Azure and others so reports this 3rd party revenue on a NET basis in their total revenue. Anthropic INCLUDES the revenue share that goes to @amazon AWS and others in their revenue so reports this 3rd party revenue on a GROSS basis in their total revenue. IMO, OpenAI taking more conservative approach that reflects the reality of the economics of these hyperscaler partnerships.

@5rb6jj7wtx @deedydas For sure - but still interesting. The fact that it is written directly means eg you could chuck autoresearch at it 👀 @deedydas

@deedydas @cclark Right, but surely it read existing engines, I’d question the use of the word novel here

I just "vibecoded" a Chess master (~2250 ELO) from scratch that runs locally on a Mac in Rust. I used to play chess semi-competitively, and I'm flabbergasted that you can just speak a 98% percentile chess engine into existence.

thanks to coding agents it's never been easier to get started, and never been harder to get finished.

In a moment of frustration, I banned my 8-year-old from saying “I’m bored” and he now has to say “time to figure out a new activity” and it’s been weirdly effective. Also I’ve threatened to take away dessert if he says it. That also is def part of the success recipe.

Looks great! I have not read the Chinmayananda version, but I have the Easwaran translation of the Gita and it seems more approachable. Not sure if it's public domain though. Chinmayananda: What did the sons of Pandu and also my people do when, desirous to fight, they assembled together on the holy plain of Kurukshetra, O Sanjaya? Easwaran: O Sanjaya, tell me what happened at Kurukshetra, the field of dharma, where my family and the Pandavas gathered to fight.

These three books outsold every novel, outlasted every empire, and are the calling for 70% of the world. But you can't find a good copy on the internet. And you can't take them offline with you. And you can't read them on a plane. So I made a little thing:

If the east wing ballroom had been constructed by Obama, what kind of impact would that have had on the plot of White House Down?

Bullish on the Workdays of the world. Well-structured line-of-business software and effective systems of record are not going anywhere. Good data structures, with mature APIs, are the perfect systems for agents to interact with, and not create a mess in their wake. AI doesn't need to live inside the tool, and building properly governed enterprise software is not trivial.

With models training on other model outputs — feels like only a matter of time — before model outputs — are primarily — dominated — by ———

Whoa! Exciting! Congrats to all involved and look forward to using the resulting products!

@thdxr @pingToven @charlesdotai @alexatallah Credit where credit is due - I think you noticed and did something about tool call variability between providers before anyone (including us) understood it well. We wouldn’t be at this point without that work. Thank you!

@pingToven @charlesdotai @cclark @alexatallah nice

So excited for this to be live. months and months and months of work. huge shoutout to my team, especially @charlesdotai for all the infra work, @cclark for the original exacto work, @alexatallah for the support, and many others.

need a word for 'mindshare' but for llms & agents. 'weightshare'?