Simon FL (@simonfl) header
Simon FL (@simonfl) profile picture

Husband to @pearlsesq, Software engineer @databricks, French Canadian (i.e. likes poutine and hockey), previously @stripe, @SlackHQ, @Foursquare, @Google

New York, NY
Joined March 2007
Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)Photo by Simon FL (@simonfl)
Simon FL retweeted
This is one of those unintuitive things. Agents that cook longer are often worse. Genie just gets to the results faster. Ontology will be key to getting these agents the context they need to get the answers right quickly.
We found that improving data agent quality also improves efficiency. Sometimes more is less: agents that take long, exploratory random walks are often less likely to arrive at the right answer. We put Genie Code head-to-head against three leading general-purpose coding agents on 400+ real user data tasks. Result below 👇
17
27
1
161
28,975
We just published OfficeQA Pro - a set of 133 challenging questions from the original OfficeQA benchmark. Even the best frontier agents still struggle on OfficeQA Pro with common issues stemming from errors in parsing, retrieval, and visual reasoning.
1
8
25
2,412
Most AI benchmarks test reasoning in isolation. Real enterprise tasks require grounded reasoning: 1️⃣ Find the right documents 2️⃣ Extract the right values 3️⃣ Perform analyses OfficeQA Pro evaluates this end-to-end. Frontier agents still score <50%. 🧵Paper & details below!
7
27
10
110
45,224
Today we’re releasing OfficeQA — a new benchmark for end-to-end grounded reasoning that reflects the real work enterprises need AI agents to do. More details below 👇
Today we’re introducing OfficeQA, a new benchmark grounded in ~89,000 pages of U.S. Treasury Bulletins that reflects the complex, document-heavy tasks enterprises actually face. Unlike existing benchmarks, OfficeQA measures economically valuable, real-world reasoning: parsing dense tables, navigating scanned PDFs, and retrieving facts across decades of documents. Even strong agents reach only ~45% accuracy, showing how far the field has to go. The benchmark is now open to the community, and the Databricks Grounded Reasoning Cup in Spring 2026 will challenge teams to push these capabilities forward. databricks.com/blog/introduc…
4
18
2
41
9,091
Why is it "two buck chuck" and not "two bucks chuck"? Is this like the "maple leafs" vs "maple leaves"? English plurals are insane
1
3
194
Since joining @databricks, our research team has been hard at work on Agent Bricks, a new product that helps enterprises develop state-of-the-art domain-specific agents. We are now releasing a research blog about Agent Learning from Human Feedback (ALHF) databricks.com/blog/agent-le…
2
20
3
101
9,888
RLVR isn't just for math and coding! At @databricks, it's impacting products and users across domains. One example: SQL Q&A. We hit the top of the BIRD single-model single-generation leaderboard with our standard TAO+RLVR recipe - the one rolling out in our Agent Bricks product.
3
15
6
107
23,298
Hey @minimax_ai, I'm trying to serve M1-80k on vLLM. Your docs say "a server with 8 H800s can process inputs up to 2 million tokens" but then recommend --max_model_len 4096. What settings did you use for 2M tokens? I'm trying this on 8 H100s.
1
6
1,425
Simon FL retweeted
DONDA is the new FAANG Deepmind Open AI Nvidia Databricks Anthropic
106
381
74
2,509
352,395
First day on the job at @databricks and we're already making some big moves. Exciting times ahead!
Big news: we've agreed to acquire @MosaicML, a leading generative AI platform. I couldn’t be more excited to join forces once the deal closes. databricks.com/mosaic-news
2
346
I feel like today's war on cars is so much tamer than it was 80 years ago
4
341
Now this is AI thought leadership at its finest. A+. No notes
I don't mean to make a bad taste joke, but pronouncing GPT-4 in quasi-French ("gé pé té for") sounds *very* awkward.
3
287
Replying to @GergelyOrosz
I always keep in mind the following: When you talk to someone you have the responsibility to do your best to adapt to the other person skill level. If you don't then you are disrespecting the other person and you are losing your time.
1
17
2
92
48,508
That liberating NYC feeling when scaffolding is finally removed
2
11
440
There's nothing quite like seeing "thought leaders" write about a topic you're very familiar with to disabuse you of the notion that they have any idea what they're talking about.
9
571
Who needs cable anyway?
1
11
702
The team that shitposts together ships together.
9
413
A thousand times this. Hell, some of the best people I've worked with barely even have a LinkedIn profile.
Replying to @pulidili
High competence will attract the right network, automatically. Brand (if by that you mean public brand) is highly overrated for high paying professions such as prod mgmt, eng, design, data sci, etc. Majority of highly competent people have relatively small social media following.
1
1
373
How does one check the "internal temperature" or fried rice???
212
Marathon (1994)
9
63
2
641
72,523