Howe Wang (@WangUWS) header
Howe Wang (@WangUWS) profile picture

Enterprise AI engineering & deployment @MistralAI | Views my own

New York
Joined August 2021
Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)Photo by Howe Wang (@WangUWS)
You are warned by NYC gov
149
"Within @databricks , 92% of prompts were going to too large of a model. That translates into millions of dollars a year wasted on what could have gone to a more right size model for the problem." yeah model routing is a real need. #DAAIWTT
7
1
41
3,156
Howe Wang retweeted
To celebrate Mistral Vibe + Le Chonk becoming the world’s best open model-harness duo, we’re making Vibe free in Europe for the first 3 months. (Yes, it’s actually better than GLM / DeepSeek at coding). Use code LECHONK via link in 🧵. Limited time, first-come, first-served.
259
356
147
3,797
346,944
Howe Wang retweeted
PSA: Each year SEC brings ~35 insider trading cases, and most SEC insider trading cases are for < $200K in profits.
11
22
8
1,506
115,110
holy shit
96
543
38
12,913
294,303
Howe Wang retweeted
Maybe it’s about the friends we funded along the way.
5
35
5
222
37,121
Saba Capital's founder Boaz Weinstein on why you shouldn't pay a hedge fund for beta: "The platforms like Millennium don't carry that beta." "Beta is accessible through Vanguard for nothing." "So why someone would want to pay a manager and pay them incentive fees and management fees to have exposure to the S&P seems like lunacy." "Hedge funds are supposed to be delivering alpha. Beta's so cheap, you shouldn't be paying for beta."
4
36
5
372
86,699
Howe Wang retweeted
Replying to @RamaswmySridhar
12
14
4
422
36,450
Howe Wang retweeted
The Artificial Analysis Intelligence Index is currently made up of 9 benchmarks with arbitrary weights. I wanted to know how these weights varied over time. They vary quite a bit!
47
13
7
186
22,633
8
567
24
4,934
145,407
ANTHROPIC'S IPO VALUES IT AT UP TO $1.02 TRILLION, BARELY ABOVE ITS $965B PRIVATE VALUATION
16
27
24
641
100,438
Cómo me imagino a mí mismo dentro de la historia cada vez que leo un libro
95
16,346
263
125,227
1,356,495
Howe Wang retweeted
Replying to @HamelHusain
multi-vender approach makes more sense at scale. or byok and token disti from hyperscaler with spending commits and discounts. most co internal token spend should far outweigh external product token spend.
1
1
3
244
Never felt closer to a fundamental change in how we build retrievers, and a different kind of memory for agents. ARI is basically test-time training for retrievers, where gradients derived from a reasoning model's preferences teach the retriever where to look *at search time*. Lots more to do here to get oblique search queries as manageable as non-oblique ones, and that will include first figuring out sub-linear (late-interaction-esque) search. But the path ahead has never been clearer. Read @dianetc_'s blog!
Is there a way to scale a reasoning model’s judgment for searching a massive corpus? We introduce the Amortized Reasoning Index (ARI), a new kind of retriever that updates its weights during search, using gradients derived at inference time from the ranking preferences of a reasoning model. Every session, current agents spend lots of reasoning deciding which search results matter, and then throw all of that judgment away upon compaction. ARI is our attempt to “amortize” that judgment, so that reasoning spent on past queries keeps improving search on future ones. With no relevance labels, ARI takes a 400M cross-encoder that starts out weak and grows its Recall@100 on held-out queries by up to 7×, ending above Qwen3-Embedding-8B on various oblique/reasoning-intensive search tasks. Read our blogpost: dianetc.com/musings/ari/ Preliminary results and discussions are below!
22
36
3
311
18,351
Buenos días de mi familia a la suya
131
Howe Wang retweeted
everyone is a data scientist now. learn to eval.
We just released a free course on Claude Academy: Introduction to Evals. If you have a working prompt but no way to measure it or are just curious about how to get more structured feedback from your prompts, this short four-lesson course will walk you through it.
3
11
219
11,799
Most call themselves market neutral.
Comprehensive list of hedge funds blowing up this week, getting large redemptions, or shutting down PMs: - Axebrook - Walleye - Engineers Gate - Lone Pine - Millennium internal pods - Verition pods - Samlyn Letter from Walleye closing internal funds below
141