Ben Scharfstein
@benscharfstein
Followers
2K
Following
9K
Media
107
Statuses
2K
started a few companies. built ai @scale_ai, @google, applied math @harvard
San Francisco, CA
Joined August 2011
Incredible result and people underestimate how important speed is. On Google Search, latency launches usually moved DAU more than ranking launches
Today we’re releasing SWE-1.5, our fast agent model. It achieves near-SOTA coding performance while setting a new standard for speed. Now available in @windsurf.
0
0
2
Wemby and Ohtani the GOAT candidates that’s Brady, Tiger, and Lebron passed the torch to and I (and ESPN) love it
0
0
1
Pretty much everyday I believe this more and more
Context is the product
0
0
2
Sometimes I meet a team and think they must be reading my mind (or at least my meeting transcripts). That was my feeling when I met Yash, Rhythm and Linden. And then when they told me about their vision for Specific Intelligence it opened up an entire new way of thinking.
Generalists are useful, but it’s not enough to be smart. Advances come from specialists, whether human or machine. To have an edge, agents need specific expertise, within specific companies, built on models trained on specific data. We call this Specific Intelligence. It's
2
0
5
What’s the best platform these days to test 10 different prompts against the same test set? Needs to support images and be easy for non-technical users to iterate on prompts
0
0
1
Building and selling are the two things with unlimited demand as costs go to zero. Most other things have a cap — I only need so much accounting/legal/finance work to be done.
1
0
3
We’ve raised a $20M Series A from @IndexVentures, @BainCapVC, and @eladgil, with participation from @sequoia, @CooleyLLP, and @patrickc It took us 173 days to review our first 1,000 contracts. Now, we do this every three weeks. Even as we’ve scaled, we’ve kept our signature
105
76
671
The best tools to give coding agents are usually just the sa,e tools humans need to write code. Search, terminal, error logs etc. One of those obvious tools is using the app after you build it and yet I hadn't seen any agents doing this until I just tried @Replit. Bravo.
2
0
2
This also seems like the wrong conclusion. IMO A/B tests are important! Evals are important! Loss functions are important! Product principles are important! Vibes are important! There is no one answer to building great products, it’s a multi front war
This is the exact opposite conclusion to draw from the acquisition. AI enables you to build much more dynamic products that evolve faster (and even automatically). The foundation underlying that is good evals. A/B tests are officially the way of the past.
1
1
9
One of the most fascinating research agendas I’ve seen. Colloquially people using LLMs refer to them having world models because they seem to generalize well on many tasks. Keyon and his collaborators show they don’t in ways that are nuanced but important for practitioners.
Here's a video I made that goes over methods we've worked on for evaluating world models. Thank you @srush_nlp for the opportunity!
0
1
13
The thing that made Snowflake work was the separation of storage and compute which allowed people to scale analytics jobs horizontally. Agents allow for MapReduce for knowledge work. Tasks that take one person 100 hours can now take 1,000 agents 6 minutes.
2
1
7
Best kept secret in SV 🤫
The enterprise AI switch has flipped! I loved jamming w/ @benscharfstein who is as sharp as they come on building complex AI for the enterprise. He runs product for one of the best kept secrets in SV - the @scale_AI Enterprise Applications Business - which has been a remarkable
0
1
9
“I think the right way to think about foundation labs is that they're like movie studios.” Each model is a blockbuster: huge budget, short lifetime, then the sequel. And the franchise name endures. w/ @benscharfstein, Scale AI
14
10
117
Gotta do the schlep!
"Your job is not to build a product. Your job is to solve a problem." Most companies avoid the messy, unsexy work. But that’s usually where the biggest opportunities live. Do the work others won't. – @benscharfstein, Scale AI
3
0
26
The enterprise AI switch has flipped! I loved jamming w/ @benscharfstein who is as sharp as they come on building complex AI for the enterprise. He runs product for one of the best kept secrets in SV - the @scale_AI Enterprise Applications Business - which has been a remarkable
13
18
116
Knew we’d get along since I first read Joe’s post, “Trading Margin for Moat” —thanks for having me on to ramble!
The enterprise AI switch has flipped! I loved jamming w/ @benscharfstein who is as sharp as they come on building complex AI for the enterprise. He runs product for one of the best kept secrets in SV - the @scale_AI Enterprise Applications Business - which has been a remarkable
0
1
8
is it too much to ask for a full frame, interchangeable lens camera with phone quality AI onboard and a reliable smartphone connection? @marclevoy @alexschiffhauer can one of you partner with Sony/Nikon/Canon? Please and thank you
0
0
2
Behind the scenes from the first pixel launch event 10 years ago
Behind the scenes fun getting ready for #MadebyGoogle with @jimmyfallon. At our first MBG in 2016, I said that innovation lies at the intersection of hardware, software, and AI. Almost 10 generations later, we're excited to show you how AI continues to transform our Pixel family
0
0
4
This is the way. Have had (vibe)coding take home at Scale since February for our PM team. It takes time to calibrate to the interview format but once you do it’s high signal
We are adding a coding section to all of our Product Managers interviews at @Shopify. We'll start with APM interviews. We expect candidates to build a prototype of the product they suggested in the case interview. There is no excuse for PMs not building prototypes.
0
0
4
These types of product nuances in ChatGPT make a big difference. One of my biggest complaints about Gemini is that it thinks too long for simple questions.
0
0
1