Pinned Tweet
Some news!
We’re joining forces with @RivetTax.
Over the last two years running Lobby, we launched four products at the cutting edge of AI, all focused on helping teams customize and fine-tune these models to make them go from good to great
What excited me about joining Rivet was the idea of putting our code into the hands of hundreds of clients and tax preparers on Day 1 – tax season is right around the corner, we can’t wait to show you what we’ve been working on
PS. My mom is actually a CPA in Georgia – peaches don’t fall far from the tree, after all
It’s Day 329 at Rivet, but today is just the beginning
Proud to announce – we've acquired @LobbySo to power the future of AI-empowered tax prep, with Lobby's founder @WiLSONSACCOUNT joining Rivet as our founding engineer to drive our AI efforts internally
Tax prep is both art and science. The art is in interpreting IRS guidance and guiding clients through their own interpretations of their financial realities.
The science is a set of tasks that need to be completed to furnish a completed tax return to a client, most of which involve finding some specific string of numbers or letters, a needle in a haystack of emails, PDFs, Zoom transcripts, and Slack conversations.
The art of tax preparation cannot and should not be automated. But the tedium of trawling 1000s of pages of PDFs for a specific cost basis for a secondary sale, or thumbing through a transcript for the name of a subsidiary entity mentioned once, that absolutely should be. The meat of tax preparation isn’t just converting a single W-2 or K-1 into a data packet that Drake or CCH can interpret, but knowing where to find needles in client-provided haystacks.
Wilson and the rest of the Lobby team spent the last three years focused on reliably pulling specific answers from reams of unstructured data. And just in the last month of working together, we know we’ve made the right call having Wilson join the team.
Personally, I’m just excited for next tax season when we can put all of his & the rest of the team’s hard work all to good use! 🥂
wilsoNYC retweeted
83(b). 1120. 5471. 26.2.
Only one of them was optional.
@RivetTax’s own Ana Mía, running NYC this morning 🏃
wilsoNYC retweeted
Karpathy 开炮了（前 OpenAI、前特斯拉 AI 负责人）：
"Agent 不是魔法，是大规模蒸馏。你模型 99.99% 的容量，都浪费在它根本不需要的数据上。"
16 分钟对谈，他给出了真正有效的技术栈：小模型、正确的工具、一个闭环。
大模型 → 小模型 → 正确工具 → 闭环 → 自我改进的 Agent
别再为更大的模型交税了，让小模型变危险的，是闭环。
这 16 分钟，比今年任何 500 美元的 AI 课都值。看完再读"Harness, Loops, Graphs"——闭环就是那副马具 ↓
“scarce DRAM should sit in my data center” is a fascinating conclusion for the guy selling data-center tokens.
the efficient frontier is: route the 80% of easy work to silicon people already own, save the H100s for the hard stuff
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
wilsoNYC retweeted
turns out you can get frontier performance by having a swarm of tiny models working on the problem together.
great work from the research team on this one :)
cohere.com/blog/lessons-from…
Announcing new research on scaling multi-agent teams for long horizon software engineering tasks. 🤖🧩📈
Work led by @keenansamway, @mrdanieldsouza, Omar Darwiche Domingues, @mziizm, @yichern_tan, and @nouhadziri
Read the paper: cohere.link/tStRYJS
Check out the blog post: cohere.link/0TzgKoy
wilsoNYC retweeted
Restaurants are dead.
Chefs are dead.
I just bought an airfryer and I can cook anything.
And recipes are on the internet for free!!
wilsoNYC retweeted
We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case).
I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves.
We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them:
"/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started"
I imagine we'll have some backend services that could be replaced entirely by this new architecture.
cancelling all investor calls this morning i’m busy
wilsoNYC retweeted
Pro software perf tip:
Doing nothing is wayyyyy faster than doing something.
planetscale.com/blog/designi…
wilsoNYC retweeted
I experienced the Silicon Valley scene
“You mean my Aviato?”
“There is only one Aviato”
❤️😊
——
An angel investor lead the greeting part of our sit down with appreciation of our prior work. I felt special, it was hot
Everyone is flaming this guy but I have seen founder girls on hinge who use the same rhetoric
Part of dating in SF is you have to want to be the next bill+hillary
i’m looking for a VC girlfriend. having talked to 100s of VCs and founders, here’s why i’m 100% serious that this is a win-win:
- i’m raising my series A for @KoyalAI in a few months. If you backchannel me, we’ll both get to Series B faster.
- i’ll be your scout for founders before they get to YC/a16z. My DMs are full of ambitious people starting companies.
- i can take you to nice restaurants and you can expense it as “diligence.”
- you probably take better care of yourself than most founders. You can womansplain biohacking to me
- most of my friends are founders who hate VCs, so you’ll get a front-row seat to the gossip. I expect the same privileges in return.
- we’ll never have to argue about who pays for coffee chats. Your fund does.
- if the relationship works out, great. If it doesn’t, we still generated deal flow for our respective organizations.
- we probably won’t get bored talking about tech, but can also go to burning man.
please DM or tag relevant friends.
23+ years old, seed and above, serious inquiries only, thx.
again, not a joke. X do your thing.
wilsoNYC retweeted
Matt, thanks for helping me find so many soon to be customers.
Automated local model training incoming
This mindset destroyed my health as a solo founder and I am still recovering
Do not believe the selection bias
wilsoNYC retweeted
This will be a known very important problem for banks and fintech’s in the next six months.
It’s a known issue by enough of the market that Socure has grown to fill the need, but there is going to be fire and ash in loan application funnels that don’t have good Sybil Resistence
Bot detection & human verification will be one of the most urgent demands for businesses over the coming years. Agent swarms will suffocate every website and form; small companies and government websites are most vulnerable.
There is a huge gap in the market for this right now. When we looked at what offerings were in the market to use at X, there was not a single company that brought together all the latest technologies so we had to do it all in-house.