Detaining a PhD student on her way back to Princeton from presenting scientific work at a conference is not how we get the best and brightest scientists in the world to want to come to the US. I hope this is resolved quickly! More details:
dailyprincetonian.com/articl…
I'm only a tiny investor but unreasonably proud of @lancedb. It has the originality, camaraderie and technical excellence of OG Cloudera so it's more or less destined to change the world.
Just ask @changhiskhan @jonhsieh @colinrmccabe @jordan_volz @hcoyote @Farmertodd3 & more
Our customers are building AI that generates video, drives cars, and helps robots understand the world. We just raised a $57.5M Series B to help them do that faster, led by @BessemerVP and @iendeavors .
Years ago, I helped write pandas for a world where data just meant numbers of strings. Later, running ML engineering at Tubi, I saw the limits of that world up close. The data that matters for AI now is video, images and sensor logs, and the tools we all grew up with were never built for it. @eddyxu and I started @lancedb because we lived that problem and couldn't stop thinking about it.
Nowhere is the shift clearer right now than in physical AI. Robots and autonomous vehicles learn from enormous volumes of video and sensor data, and the teams building them were losing weeks to data wrangling before every training run.
Teams like @runwayml and @nvidia Cosmos curate millions of hours of video using Lance and is now training directly from it. Earlier this month at Reverie (reveriesummit.com), our first summit, the topic robotics and autonomy teams kept coming back to was the data flywheel, and how to make it spin faster.
That is what this round is for. We're building the multimodal lakehouse so that a researcher's only limit is their ideas, not their infrastructure.
Since our Series A, revenue has grown more than 10x year over year, and open source downloads have grown 6x in the past year. More importantly, we've a team that I am proud to be part of every single day, customers who trusted us with their hardest workloads, and a community of contributors and maintainers who make the project better every day.
To our investors: thank you. @BessemerVP and @iendeavors for leading, and @Theoryvc, @_DimensionCap , @swift_vc , @CRV and @ycombinator for backing us again and again. You believed in what this could become, and we don't take it for granted.
We're still in the early innings. If you want to build the data foundation for the next decade of AI, we're hiring.
lancedb.com/blog/lancedb-ser…
We've been refining the agent experience for our cloud for nearly a year now and it really shows in the results.
We regularly benchmark our competition and some of them... have work to do.
Big congrats to @nutlope for shepherding @togethercompute towards a stellar agentX.
We need a lot more inspo like this. Can't wait for more "truly native" AI applications along these lines.
AI interfaces nowadays are still pretty much turn-based.
In my PhD thesis, I thought about what interaction would look like if we could go beyond taking turns.
One example is intra-turn interaction:
1
5
595
Charles Zedlewski retweeted
AI interfaces nowadays are still pretty much turn-based.
In my PhD thesis, I thought about what interaction would look like if we could go beyond taking turns.
One example is intra-turn interaction:
This experience has been so very important to me.
My brother is one of the best to *ever* do the thing.
Building a company to capture his imagination for 10+ years is a feat greater than CUDA. And @JensenHuang did it with such class He so much more than a CEO or a scandal.
I have a particularly nasty case of cancer. When I was hospitalized again at my lowest point last year, I told Jensen. He immediately reached out to contacts at Stanford, who were able to cut through red tape, and get me into a clinical trial for which I've turned out to be a perfect fit. I think everyone can have their own views about recent events. But any claims that Jensen did not really care about his staff is simply inaccurate. I, personally, owe him more than I can express.
I think the best tribute we can pay to Jensen's legacy is to drive NVIDIA's continued success and push the whole industry forward, by showing we've learned the lessons from years of working here. We will compare ourselves against Speed of Light. The mission will be the boss. And we will not be scared to take smart, bold risks and fail fast with ruthless intellectual honesty.
But if there is one management story I remember the most, it's a particular Jensen grilling from my time in autonomous vehicles that stayed with me. Luckily I was not in the hot seat for this one. Team A didn't want to use Team B's framework, because they thought it was too slow. Jensen was surprised - it's just applying a bunch of optimized CUDA kernels in sequence, how could it be slow?
So he asked the executive from team B to draw the entire data flow sequence from their framework on the whiteboard and highlight every memcpy. Sure enough, there were a bunch of unnecessary memcpy's surrounding each pipeline stage. Everyone in the room could see why it wasn't getting to SoL and a path to improve things. No arguing about fiefdoms or appeals to authority - the data got us all on the same page.
The lesson was not (just) that unnecessary memory copying is bad. It's that executives should not be afraid to probe into details if something doesn't add up, and leaders need to be technical enough and honest enough to engage in real time and not turn every meeting into "I'll get back to you in a week." If that means counting memcpy's with a billionaire CEO, so be it.
My best executive meetings at NVIDIA have been when I'm fully prepared to present every detail of my team's work, and my worst failures have come when I realized I couldn't go deep enough to really solve the problem.
But the thing that gives me confidence is that Jensen was never the only one to live this philosophy. Anyone who has presented to an SVP at NVIDIA knows they are at risk for getting grilled on fundamental constraints and technical details. I have been part of amazing conversations with both recent and old timer leaders where we were able to get to the core of a problem in an hour rather than a month, because we had the right people in the room and were not afraid to go deeper.
The company is filled with supremely talented people who have been trained for 3 decades to work at the highest level, and you can see it in what we produce every day. I'm so proud to be an NVIDIAn, and I can't wait to show off what we ship in the coming months and years.
Never thought a MTV Real World cast member would stand between us and planes colliding in the sky...
2
280
There's a ton of potential for Apple compute for personal agents, production apps and model training.
Apple silicon needs a unique operating model & supply chain and no one has it sorted better than @petervoell .
Agentic AI runs best on Apple platforms. Accessing Apple’s hardware at scale has been hard, until now. We’re making macOS and Apple silicon available as elastic compute resources, optimized for AI research & deployment.
We have one of the largest fleets in the world, support the most demanding workloads (massive 512GB M5 Ultra clusters, anyone?), and we’re just getting started.
Wouldn’t be here without @CRV (@brittwalker_), @conviction (@pranavreddy), @chadbyers, @offlineventures (@davemorin), and others (including many inside @Apple) who believed in us early and remain incredible partners. We seek infinite games; the play continues.
blog.mountthor.com/a
3
10
755
Have we reached peak tech brand aura farming? Every startup web site and pitch deck these days:
3
233
Charles Zedlewski retweeted
open models are 11 months behind
open models are 8 months behind
open models are 5 months behind
open models are 1 month behind
It all depends on what you're measuring
Here are some ranges for different benchmarks
It has been a blast collaborating with @JulienVerlaguet on @skiplabs. Congratulations @MistralAI on your très bon goût!
@JulienVerlaguet has some inspiring ideas for how programming language design will accelerate model development so watch this space.
🚨 EXCLU
@MistralAI rachète SkipLabs, une startup française qui rend la génération de code par IA plus rapide et plus fiable.
C’est sa 5ème acquisition en 9 mois, après Koyeb, Emmi AI, Pimento et New Material.
Portée par sa levée de 3 milliards d’euros, la pépite française élargit son terrain de jeu face aux géants américains et chinois.
Jusqu’où ira Mistral AI ?
🔔 Retrouvez toute l'actualité sur Maddyness.com
6
414
With the cost of compute only climbing, freeing your data from any single cloud is a wise move.
Really happy to see @TigrisData will continue to grow it's global footprint as part of @Dropbox.
Congratulations to @ovaistariq on building a great product and a great company!
Charles Zedlewski retweeted
Introducing Together Compiler: agents that continually optimize inference for production traffic.
Serving trillions of tokens/day. Up to 2.6x faster than leading InferenceX configs, up to 50% faster on customer workloads.
Congrats @reyna_abhyankar @dylan__lim @simran_s_arora and the team!
Introducing the Together Compiler: continual compilation for AI inference! I’m super proud to finally share this work from our research team! ✨
Today’s inference is “Polymorphic” or shape-shifting, it doesn’t sit still. On OpenRouter, 80% of traffic goes to models released in the last three months and the top ten most popular open models each use a different architecture. Our customers constantly change their usage behaviors. Together Compiler leverages agents to continually optimize our deployments immediately as these changes occur.
Together Compiler research highlights include:
1. 252x better throughput: TKCC lets agents try 252x more optimization experiments per hour compared to prior agent inference optimization systems.
2. State-of-the-art TPS across InferenceX workloads we tried: TKCC outperforms the best InferenceX engine across all tested settings in just 24 hours & 2 nodes per compile. Including 1111 TPS on MiniMax M3.1 B200 and 1205 TPS on DeepSeek v4.1 Flash GB300 at batch size 1. NVIDIA, AMD, Trainium2.
3. Continuous recompilation: TKCC currently serves many trillions of production tokens per day and adapts to traffic shifts within just a few hours of recompilation.
4. Bitter lessons: TKCC’s design builds on 2+ years of building frameworks to accelerate AI Inference, starting from ThunderKittens and Megakernels. Compilers have always used abstractions and IRs, but we show that agentic compilers now reach higher performance ceilings without abstractions.
From an incredible team @reyna_abhyankar @dylan__lim @realDanFu @togethercompute!!
Charles Zedlewski retweeted
The last two elections have made me hopeful. Maybe that’s weird but it’s true. If you do dumb shit the median voter does not like that and will kick you out. This great American experiment of ours seems to be working fine. Onward.
Charles Zedlewski retweeted
Europe’s largest private company just made open models its default for coding.
@Revolut: ~2,000 engineers, 1/10th the cost, 70% faster lead time.
Proud to be part of the infrastructure powering open models in production 🚀.
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
Charles Zedlewski retweeted
Three podcasts I really enjoyed recently
1. Cliff Sosin - Owning the Complements to Intelligence. This was really stellar. Cliff makes a compelling case that intelligence is being commoditized, and leads a conversation to consider what becomes valuable and scare in that scenario. I don’t fully agree with him and still found it be incredibly insightful and compelling
podcasts.apple.com/us/podcas…
2. Dr. Sarah Paine: Why The World Is Turning Away From America. I found this to be quite depressing but worth listening to. Dr. Paine is always a great listen, but it’s not exactly fun to sit with how catastrophically Trump’s second term has eroded our standing as a nation, destabilized the world and materially degraded the future outlook for our children. Required listening nonetheless
podcasts.apple.com/us/podcas…
3. Robert Caro – The master biographer reflects at 91. I am conflicted. Obviously this is great. But I don’t like that he is being distracted from finishing the last book. Anyways. It’s here so listen to it. It’s great. Read all his books if you haven’t. What more is there to say?
dwarkesh.com/p/robert-caro
Charles Zedlewski retweeted
one of the coolest things we've open sourced is our read-through cache for object storage!
we've also added support to cache exact byte ranges, helpful for reading small ranges from large objects
github.com/s2-streamstore/ca…
Charles Zedlewski retweeted
Neat case study on AI for systems development. It turns out you just need (0) smart humans to understand and communicate solutions, (1) good metrics and cost models, (2) super basic swarms of agents to write code and tests to optimize these metrics
On a side note: now that AI has gotten so good, finally systems PhD students & academics can, and should, build extremely ambitious systems that cover big surface areas, and evolve these systems out in the open over time (1+ year)
I built bbtree, the foundation of databases and filesystems, and the fastest concurrent btree.
Code: github.com/XiangpengHao/bbtr…
Blog post & demo: bbtree.xiangpeng.systems/
When the AI startup fails and has to give back its compute: youtube.com/watch?v=vEB6aQkA…
Heard from a neolab friend:
“Even if we fail to train the best model, we locked in GPU capacity early. H100/B200/B300 rental prices have since gone up 2–3x. We can just resell the compute and print money.”
OK, so that’s the business model.
2
790
Charles Zedlewski retweeted
Two weeks after launch, Zephon v0.2 is here.
It now reads straight from LanceDB tables, runs on Modal and on setups without a shared filesystem, and redistributes mixture weights for you.
Same promise as before: identical batches whether you train on 8 GPUs or 16.
uv pip install -U zephon
Zephon launched two weeks ago. v0.2 is out today.
→ Use data stored in your @lancedb tables
→ Run on @modal & setups w/o a shared FS
→ Mixture weight redistribution
→ More zero-copy ops + reliability fixes
w/ @josh_wills & the @datologyai team
uv pip install -U zephon



























