Johnna Liu
@OvercookedJoJo
verifiable harness for self-improving agents @sponsiolabs , cs dropout @Cornell , exTMT IB @GoldmanSachs
387 Following 2.3K Followers
thanks Astra @OpenAI.
it’s been a while since I last traded, but 30% of my portfolio is $SOXL and another 30% is $MU + $SNDK.
we made the same mistake especially when we are selling to end clients who are not in the sf bubble where startups just cross sell to boost unreliable arr
We did $100k in revenue after realizing that our customers hate software.
Our first product was an AI accounting tool for CFOs. I did 35 sales calls and got $0.
So we switched to being a service.
The pitch became: "We're an outsourced accounting firm that's faster/cheaper with AI"
Then our calls started to convert.
As founders, we naturally obsess over features and tech because that’s what we do.
But REAL customers don't care. They only want their problem solved (bonus points if they don't have to use the tool) in a fast, simple way.
Just sell the outcome.
Show more
we made the same mistake especially when trying to sell to clients who are not in the sf bubble where most startups just cross sell to boost unreliable arr
We did $100k in revenue after realizing that our customers hate software.
Our first product was an AI accounting tool for CFOs. I did 35 sales calls and got $0.
So we switched to being a service.
The pitch became: "We're an outsourced accounting firm that's faster/cheaper with AI"
Then our calls started to convert.
As founders, we naturally obsess over features and tech because that’s what we do.
But REAL customers don't care. They only want their problem solved (bonus points if they don't have to use the tool) in a fast, simple way.
Just sell the outcome.
Show more
amazingly creative launch video
Agents f*ck up.
We raised a $2.3M pre-seed to warn you before it’s too late
working at a quant fund is like getting a fat paycheck to become a palace eunuch🤐
My former employer Citadel saw my free content on the power industry.
They sent me a cease and desist to take it down.
But I'm leaving it up. It's all based on publicly available information, and it's becoming increasingly important for everyone to understand power markets.
Show more
one of the reasons why Anthropic narrowed its IPO plan to late September / early October
last time it was "by the end of this year"
We can finally talk about it:
We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company.
We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried.
Show more
holyyyyyyy
I am excited to announce that I, along with my long-time colleagues (@JeffDean, @OriolVinyalsML, @quocleix), are founding Discovery Loop (@DiscoLoopAI), a public benefit corporation. We aim to automate the process of discovery and progress in machine learning, science, and engineering. The four of us have helped build some of the world’s most used products and infrastructure. We’re now excited to turn our attention to this ambitious endeavor. ♾
Show more
I'm a former Citadel quant who covered power & gas.
There's constant talk about chips & memory, but power is the central bottleneck for AI.
Very few people understand it, so I'm posting a canonical primer on power pricing & data centers:
Show more
"The first thing we do, let's kill all the lawyers"
if i'm reborn with memories i am gonna do startup in ai for law and make these fucking assholes all unemployed
Software engineers won’t disappear. But how we evaluate them will change.
It used to be writing elegant, bug-free code. In the future it will be architecting agents that work across models.
Huge conflict of interest. A cracked harness is tons of engineering/dirty work plus some research.
Frontier labs have to spend their talents on internal harnesses. Making their own model stronger offers way higher ROI when the model race never ends.
The neutral harness that makes users’ multi-agent, multi-model workflows work better has to come from startups with no model to favor.
Show more
想给梁圣提一个小小的建议：
可以考虑尽快将 DeepSeek Harness 从 DeepSeek 体系中独立出来，单独成立公司，双方不再保持直接隶属关系，而是通过参股方式继续协同。
同时，产品名称也不应再局限于 DeepSeek Harness，而应重新定位为面向所有主流 LLM 的通用 Harness 平台。
这样做的好处是：Harness 不会被单一模型品牌限制发展空间，DeepSeek 也不必长期承担工具链业务带来的组织和战略负担。双方各自聚焦，又能通过股权和生态继续共享成长。
@deepseek_ai 如果方便的话帮忙转达一下，谢谢。
Show more
@patrickc just announced free Stripe Atlas incorporation for all @ycombinator Startup School attendees.
😂I wished we incorporated our entity later
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
Show more
Huge congrats to @runta! @guanlan is my role model as infra founder🫡. Runta team is also one of the deepest infra teams I know!
Harnesses matter a lot for efficient self-improvement, and formal verification will be an essential piece.
Sharing an exciting poster I saw today at @icmlconf: Their ablations are pretty convincing. Once you swap verification out for weaker rewards like unit tests, self-play degrades fast. The signal is too brittle and errors compound over iterations.
Show more
new post on harness engineering for AI self-improvement:
It is hard to forecast how much the future of RSI will rely on harnesses. Likely harness engineering will evolve in the direction of self-improvement and enable auto-research, and, in turn, smarter models keeps harnesses simple.
Even when many harness improvement get eventually internalized into core model, the need to specify goals and context will not disappear.
Show more
I’m at ICML in Seoul this week (July 6 to 12ish) for our paper on diffusion language models.
Also building something in agent safety and formal verification. Team has more papers in progress as well.
If you’re a researcher or founder working on similar topics, or an operator in a high-stakes industry like finance (quant, hedge funds, wealth management, insurance) deploying agents, we should talk! Feel free to DM me and let’s grab a coffee 🤗
Show more
BD is way more technical than I thought.
two takeaways from BD friends I talked to, which helped me finish B2B sales 101:
> stop aggressive selling. A banker-turned-BD friend got dinged early for being "too competitive", pushing the sale instead of showing the value he brings.
> the real skill is reading the dynamics across every party and finding the entry point where everyone wins.
Show more
Wow mom Garry liked my post immediately!!
founders treat what they built like their own babies.
that's also why, when we're picking which market to enter, we avoid the ones where the incumbents are still founder-led.
very easy to out-execute professional management, especially those who've never taken extreme risks to build something from scratch as underdogs.
but extremely hard to out-care an iron founder.
Show more
founders treat what they built like their own babies.
that's also why, when we're picking which market to enter, we avoid the ones where the incumbents are still founder-led.
very easy to out-execute professional management, especially those who've never taken extreme risks to build something from scratch as underdogs.
but extremely hard to out-care an iron founder.
Show more
