Hacker News - Newest: ""LLM" "Anthropic" "OpenAI""

Why Anthropic's IPO Prospectus Is Missing So Many Key Figures

cdrnsf — Thu, 29 Oct 2026 18:49:32 +0000

Article URL: https://www.wired.com/story/anthropic-prospectus-leaves-out-some-numbers/

Comments URL: https://news.ycombinator.com/item?id=50294005

Points: 1

# Comments: 0

AI #192: Anthropic Is Going Public

paulpauper — Thu, 29 Oct 2026 17:23:18 +0000

Article URL: https://thezvi.substack.com/p/ai-192-anthropic-is-going-public

Comments URL: https://news.ycombinator.com/item?id=50292704

Points: 1

# Comments: 0

OpenAI: We are doubling all usage limits for all Pro 100/200/500 tiers

tosh — Thu, 29 Oct 2026 17:13:33 +0000

Article URL: https://twitter.com/thsottiaux/status/2115850175868727621

Comments URL: https://news.ycombinator.com/item?id=50292524

Points: 5

# Comments: 1

Show HN: Apgar – Catch agent regressions without an LLM judge

jtkm6 — Thu, 29 Oct 2026 17:04:06 +0000

Hey HN, We build financial AI agents, and Apgar is the tool we wrote to see whether a harness change made an agent better or worse in production. Our offline evals can't cover every kind of question or session real users bring, and the LLM judge we run in Langfuse is too expensive to use on more than a small slice of traffic, so some failures get through.

Apgar is a small release health checker that sits on top of our Langfuse data. It picks up the release from the production traces, and for the first hours of each new release it compares the agent's behavior with the previous release on things like tool calls per turn, retries of the same failing call, loops, output length and cost per turn (11 metrics right now). The release page shows what moved in both directions. When something gets worse, we get a Slack message (and a webhook) saying which metric moved and where the change is concentrated, for example only on Telegram or only in long sessions. We've been running it for a while and decided to open-source it: https://github.com/oimo-ai/apgar

It's not meant to be another monitoring tool. We still use Langfuse for that. Apgar only tries to answer "is this release worse than the last one?" It works with Langfuse or with anything that sends OpenTelemetry. There's no LLM judge in it, so it's cheap enough to run on all traffic. The downside is that it can't tell that an answer is wrong if everything else about the turn looks the same. That's still what our evals are for.

It runs on Docker Compose (or Helm) with ClickHouse, and it's Apache-2.0. The Agents page can load demo data if you want to try it without connecting anything.

Curious how others deal with this. When you change your harness or prompts, how do you find out whether the agent actually got better or worse?

Comments URL: https://news.ycombinator.com/item?id=50292363

Points: 3

# Comments: 0

Ask HN: Sitting on unused OpenAI / Anthropic / AWS credits that will expire?

xiaomaijj — Thu, 29 Oct 2026 16:28:05 +0000

I heard someone can recycle unused AI quota

Comments URL: https://news.ycombinator.com/item?id=50291687

Points: 3

# Comments: 2

MCP tools should disclose LLM and external API dependencies

boredumb — Thu, 29 Oct 2026 15:44:34 +0000

Article URL: https://santurcesoftware.com/mcp-tools-should-disclose-llm-and-api-calls

Comments URL: https://news.ycombinator.com/item?id=50290905

Points: 1

# Comments: 0

LLM Voice

ibobev — Thu, 29 Oct 2026 14:57:55 +0000

Article URL: https://martinfowler.com/bliki/LLM-Voice.html

Comments URL: https://news.ycombinator.com/item?id=50290177

Points: 141

# Comments: 77

Anthropic Scion inspired source code management

wesscope — Thu, 29 Oct 2026 14:26:01 +0000

Article URL: https://flurry.codes

Comments URL: https://news.ycombinator.com/item?id=50289643

Points: 3

# Comments: 0

DeepSeek, Moonshot used OpenAI's GPT-6 Astra to train latest models

tartoran — Thu, 29 Oct 2026 14:07:47 +0000

Article URL: https://www.reuters.com/technology/deepseek-moonshot-used-openais-gpt-6-astra-train-latest-models-information-2026-10-29/

Comments URL: https://news.ycombinator.com/item?id=50289310

Points: 3

# Comments: 0

Ask HN: Why does meta suck at creating a good LLM?

freecodeio — Thu, 29 Oct 2026 12:15:21 +0000

My LLM experience is frontier models in a group chat for the best outcome. I recently added muse wonder and removed it after a few hours of use. I'd even use deepseek flash over muse for literally anything.

Comments URL: https://news.ycombinator.com/item?id=50287747

Points: 1

# Comments: 0