Sabitlenmiş Tweet
Joe Botsch
5.5K posts
Joe Botsch
@jbotsch
VC @Deviationcap, formerly @rtpvc, @OpenView & Head of Research at @MassChallenge. talking startups or complaining about Boston sports
New York, NY Katılım Temmuz 2011
2K Takip Edilen888 Takipçiler
@bznotes She let you order it in the first place?! With no judgement? Teach me!
English
Do other peoples’ wives also throw out yummy fast food you saved in the refrigerator to have over the next few days?
Asking for a friend. 🤣
English
this has the potential to radically upend my decade-long color coding system. time to get back in the lab
Google Calendar@googlecalendar
📣 New Google Calendar label colors have entered the chat!
English
@simonkalouche Yours certainly wasn’t a hype demo when we came by in New Jersey! The bots were flying!
English
Robotics Seed and Series A rounds are extremely easy because you can raise on demos. There is going to be a grave yard of robotics startups in the next 2 years when none of them have sufficient commercial traction or tech differentiation/moat to raise again.
Junyao Shi@JunyaoShi
Are robotics startups still raising money with demos? Funny how a demo can raise a Series A but can't survive a second take.
English
@chris_j_paxton Legs are just not going to be deployed in industrial environments at scale, the safety concerns are a huge blocker
English
This mermaid form factor has become extremely popular
𝐀𝐆@AGkorthos
A South Korean startup has just raised the largest Korean Series A, raising ₩155 billion, roughly US$103 million. Their wheeled FRIDAY humanoid is built for manufacturing with undisclosed PoCs underway across automotive, semiconductors and logistics. Their roadmap is explicit: • 100 units of annual production capacity in 2026 • 1,000 units in 2027 • Longer-term infrastructure for more than 10,000 annually Commercialisation is targeted for the second half of this year, with the company aiming to bring the price down toward roughly ₩100 million, or about US$67,000, through scale. The round follows a ₩17.5 billion, roughly US$13 million, seed raise in 2024. Video from Holiday robotics materials Site: holiday-robotics.com X: @holidayrobots
English
@robby_mtf @credistick @yoheinakajima We can’t leave out @tnachen either but it’s fun to mess with him.
English
An interesting litmus test for the VC industry:
Agentic teammates have potentially radical benefits for solo GPs and small partnerships, but that gets very little attention — because the tone is set by large firms.
I suspect they'll talk up the transformative potential of AI, as long as it's not in their own back yard.
Enrico Mellis@enricomellis
The strangest thing about running Animal Syndication Company solo: AI hasn't just replaced headcount - it's made n=1 a structural advantage. Managing a network this size, this way, wasn't possible before without dedicated people. Now it runs on something custom I built (graduated from @openclaw — thanks @mattfalconer_ for saving me from many vibecoding mistakes). I spoke to @AndreRetterath about this recently: doing the same inside an org with n>1 means fighting complexities & red tape I simply don't have. The tooling exists. The org structures don't yet. Someone will figure it out soon though.
English
@AngadRandhawa I mean the amount of joy from owning a Whiskey Raccoon cannot be adequately measured. The embodiment of the gift you get for the man who has everything.
English
@ashl3ysm1th @vermilionfund @DominicMadori @TechCrunch Congrats it’s so awesome having you in NYC now!
English
Today @vermilionfund closed our $25M second fund.
Still backing technical founders building for technical users in AI infra, security, and dev tools.
95% of my Fund I investors came back for Fund II.
That's the number I'm most proud of.
@DominicMadori from @TechCrunch wrote about our story
techcrunch.com/2026/07/08/sol…
English
Seeing so many of my friends in my feed now
Lists less important?
Say hi if you’re seeing this
English
@bznotes has been reading my diary again, but he’s the most wholesome VC we have on X, so it’s fine.
A fab prints money for a decade after you build it. A training run is worth zero the morning someone ships a better one. There will never be a used-weights market. It’s the same capex issue with none of the salvage.
It’s worth remembering, ARM never touched silicon and ended up in every phone. Nobody thought that was the platform bet, they thought it was Intel.
Bilal Zuberi // Red Glass Ventures@bznotes
Frontier AI labs face the same brutal economics that defined semiconductors: enormous capital for each new generation, a punishing release cadence, and inevitable boom-bust cycles that concentrate the leading edge into a handful of winners. But as with chips, the durable value will likely accrue not to the companies burning capital at the frontier, but to the ones building on top of increasingly cheap and abundant intelligence, which is where the real opportunity lies.
English
Joe Botsch retweetledi
it took 1 intern 3 months of continuous work, but eventually, a quantization method that beat every other algo in the market, including @nvidia's official modelopt
to explain why this matters, i ask for exactly 69 seconds of your attention (275 words @ avg reading speed of 238 wpm):
frontier models (like glm52) are huge (~0.8T params). as released, each parameter takes 2 bytes (bf16), so overall size is about 1.6 tb
a b200 has 180gb of memory. a node of 8 gives you 1.44 tb, barely fits weights, much less activations / kv cache
must quantize the model (reduce the size of each individual parameters) to serve. fp8 quantization means each parameter takes 1 byte (fits in 0.8 tb), fp4 takes 1/2 a byte (fits in 0.4 tb)
cutting the model to a quarter its original size is necessary for it to run a) cheap b) fast, and every lab serving models does this.
but, quantization lobotomizes the model if not done correctly (this is why you see people complain about @AnthropicAI nerfing claude or @OpenAI nerfing codex)
there are currently several algorithms (like Nvidia's official model-opt) that attempt to figure how to quantize a model with the least amount of damage.
they find the redundant layers that can be slashed, and sensitive/important layers that need to stay in full-precision.
these algo's have two drawbacks:
1) they take a long time to run
2) they quite often result in a sub-optimal configuration
for the past 3 months, a research (and, as always, waterloo) intern on our model perf team (@the_joshua_hill) came up with a new quant algorithm.
it consistently finds the optimal configuration:
a) in less time than SOTA
b) with more aggressive quant than SOTA
c) scoring higher on benchmarks than SOTA
achieving just one of the above is a feat on its own.
all three...excited for the paper to come out this week
Joshua Hill@the_joshua_hill
Some teaser results for a new quantization method we've been cooking up🧑🍳 GLM 5.2 is getting even faster
English
@JunyaoShi It’s like taking my high school football highlights and asking why I’m not in the NFL
English
Are robotics startups still raising money with demos?
Funny how a demo can raise a Series A but can't survive a second take.
English
Joe Botsch retweetledi
A very interesting data point on whether distillation gives you a competitive advantage. Sonnet 5 is just objectively worse than GLM 5.2, and is likely around a similar size. Sonnet 5 was almost certainly distilled (and not just distilled from tokens, but proper distillation from logits) from Mythos. This should give you an update on how relatively unimportant distillation is, in particular for the Chinese labs. It's just a way to warmstart the model for RL; the American frontier labs do it with human data, Chinese do it with tokens from American models, but they could get to basically exactly the same model without it, it would just be a little bit more expensive and require more human data collection coordination/aggregation
English
Please leave your bubbles. Most people think Codex is a thing to help you identify wild Pokémon
English
@AngadRandhawa It’s very weird living in the future then you step outside the tech bubble and get the culture shock
English
feel this often.
especially investing early stage in Silicon Valley where you’re inherently early to everything.
then it becomes about gauging when that Overton window is going to fly open
Kenneth Auchenberg 🛠@auchenberg
Thought: As an early-stage investor it’s very interesting to see growth investors getting excited about something you looked five years ago.
English
@auchenberg You’re experiencing the ultimate high…being right when someone else was wrong, second only to being right when @AmanKabeer11 was wrong ❤️
English
Thought: As an early-stage investor it’s very interesting to see growth investors getting excited about something you looked five years ago.
English
It’s ok to be scared of something AI does every now and again right?
Rohan Paul@rohanpaul_ai
MIT just built an AI that can control your body. It can move your fingers, make you play piano, even if you don’t know the song! AI decides the hand movement. Wrist pads send signals to your muscles, so your fingers move even if you don’t know how
English
Oratomic announcing a $300m Series A from the good folks at Spark, ARCH, Khosla, Index, Jeffy B & a host of other incredible folks.
If there is any team that is going to make quantum computers useful & ubiquitous. It is Oratomic.
Happy bag holder from day one here at Nebular.
Oratomic, Inc.@TeamOratomic
English