Leonard Tang
@leonardtang_
Followers
3K
Following
9K
Media
85
Statuses
2K
You don’t need frontier lab resources for frontier lab automated LLM evaluation. To prove this, we’re open-sourcing j1-nano and j1-micro: two absurdly tiny (600M & 1.7B parameters) but mighty reward models competitive with orders-of-magnitude larger peers. j1-nano and j1-micro
29
71
623
you know what time it is!.
are you using llms as a judge?. come check out our talk with @haizelabs on how to scale them up. the talk is this wednesday, make sure to sign up, if you can't make it we'll send you the study notes and recording.
1
0
7
Supported models (more coming soon):. > OpenAI GPT-4o Realtime.> Google Gemini Native Audio.> Amazon Nova Sonic. All unified under one spoken() call. Let us know what else you'd like spoken to support!.
3
0
3
In one call to spoken, you now have:. ✅ Input ASR.✅ Output ASR.✅ Raw output audio
1
0
4
Voice agents and audio-native models are fast, fluid, realistic, and end-to-end. You speak in. They speak back — no text in the middle. But every provider has their own messy, idiosyncratic API. spoken cleans it up!.
2
0
6
New open-source alert!. spoken: a unified abstraction over realtime speech-to-speech foundation models. Run any S2S model from OpenAI, Google, Amazon — one interface with one line of code.
8
5
78
EVALS.EVALS.EVALS.
lots of talks . 1. coding agents w/ @cline and @cognition_labs .2. retrievers and reranking w/ @lancedb .3. document processing w/ @ExtendHQ .4. scaling judges w/ @haizelabs
0
1
11
Quick walkthrough here:. > Which dove do YOU think is prettier?.
2
0
5
Let's say we wanted to compare images of the Haize dove -- one flying over NYC and the other SF. In a few lines of Verdict code, we can build a Judge agent to do exactly that!
1
0
6
Verdict systems can now judge image inputs. Score product photos. Ad creatives. UI mockups. Haize anime birds. Judge any thing for any quality—and understand why.
3
10
44
EVALS.EVALS.EVALS.
excited to invite @leonardtang_ from @haizelabs to come talk about auto-evals by scaling llm-as-a-judge workflows
3
2
39
get ready for session #3 of the NYC AI Reading Group w/ @qw3rtman @willccbb. > where: @haizelabs hq .> when: thursday 6/26 @ 7:30 pm .> what: mind the gap -- examining the self-improvement capabilities of large language models .> who: awesome people like yourself :^) . > and of.
7
5
79
RT @verrsane: @jxnlco @haizelabs I am always thinking about @leonardtang_ and appiny g llm as a judge.
0
1
0
<1 hour turnaround times . INSANE.
Today we’re introducing Crosby, a hybrid AI law firm that helps rapidly growing businesses execute faster. Contracts are connection points. They allow companies to transact with one another and create economic growth. But while every aspect of business has sped up, the way we.
1
0
14
💯. Pliny a real one.
@_opencv_ easy there—not a plant. if you had no job, CC debt, student loans to pay off, and an inbox full of offers, you might sign on for a few 15-hour contracts out of curiosity too. what you’re not seeing is the half a dozen full-time in-house offers with outrageous TC that would be.
0
0
8