Pinned Tweet
Introducing @Sable. We're building Aidan — an AI employee that can see your screen, speak naturally, and operate any interface in real time to run the sales conversations that drive revenue.
It's been a blast working together with @Nim_Ravid1 @chensterman and @itaprf towards this vision. We've brought together an insanely talent-dense team of engineers and researchers and are growing fast to meet customer demand. Companies like @Notion and @Decagon already rely on Aidan to run customer conversations at scale.
We've raised $45M led by @Sequoia and @8VC, with the support of BoxGroup, Sabrina Hahn, SVA, and Valor. Grateful for all of them. We're just getting started, and we're hiring across engineering and research. Get in touch!
Introducing Aidan, your newest AI employee
Aidan is the first computer-using AI built for realtime conversation
We’ve raised $45M from @sequoia and @8VC to bring Aidan to the world
@NotionHQ and @DecagonAI already use Aidan to run customer interactions. This is how @sablehq works:
Linda He retweeted
Introducing Catalyst.
What do you believe? How would you trade on it?
You notice data centers are popping-up everywhere, gold will skyrocket, or just know that Labubus will go out of fashion.
Catalyst researches your ideas, builds the optimal strategy, monitors events, and executes trades. Autonomously.
Linda He retweeted
One of the biggest constraints on growth has always been the ability to scale human judgment.
At Palantir, we saw firsthand why exceptional people matter in helping customers understand and deploy transformative technology. The hardest work wasn't explaining the technology. It was understanding context, reasoning through novel situations, and helping customers make better decisions. Until now, that hasn't been scalable.
That's changing.
@8vc led @sablehq's $45M raise alongside my friend @shaunmmaguire at @sequoia. I'm honored to be joining the board.
The best founders recognize platform shifts early and execute even faster. Nim Ravid and the Sable team saw where multimodal AI was heading early and built with remarkable speed.
Sable is building an AI employee that can see what a customer sees, interact with the product alongside them, and reason through complex problems in real time.
What felt like science fiction a year ago is already working in production with some of the fastest-growing AI companies in the world.
We’re excited to work alongside @Nim_Ravid1, Linda, Leon, Itamar, and the entire Sable team as they build one of the defining AI companies of this next era.
Introducing Aidan, your newest AI employee
Aidan is the first computer-using AI built for realtime conversation
We’ve raised $45M from @sequoia and @8VC to bring Aidan to the world
@NotionHQ and @DecagonAI already use Aidan to run customer interactions. This is how @sablehq works:
Linda He retweeted
Everyone knows marketing, sales, and deployment convert better when personalized to the buyer and their specific needs. Previously you could only scale this personalization via more employees, but now you can scale it with Aidan at @sablehq.
They've unlocked a level of personalization and real time engagement that couldn't exist before the latest advancements in computer use and vision models. We're excited to partner with them, and tell your front office teams to give them a try.
Introducing Aidan, your newest AI employee
Aidan is the first computer-using AI built for realtime conversation
We’ve raised $45M from @sequoia and @8VC to bring Aidan to the world
@NotionHQ and @DecagonAI already use Aidan to run customer interactions. This is how @sablehq works:
Linda He retweeted
@sequoia blog post here
So lucky to work w/ @Nim_Ravid1, @Lindaaa_He, Leon and the rest of this team!
And of course my first time being on a board w/ fellow troublemaker @JTLonsdale
sequoiacap.com/article/partn…
Linda He retweeted
I trained an LLM from scratch on pre-1900 text to see if it could come up with quantum mechanics and relativity.
While the model is too small to do meaningful reasoning, it has glimpses of intuition.
When given observations from past landmark experiments, the model can declare that “light is made up of definite quantities of energy” and even suggest that gravity and acceleration are locally equivalent.
I’m releasing the dataset + models and leave this as an open problem to the research community.
I also include what this project has taught me about intelligence in a mini essay linked below.
🧵(1/n)
Linda He retweeted
Dario: “The biggest thing to watch is this issue of AI systems building AI systems.”
Today, Autoscience is announcing $14M in funding by General Catalyst, Perplexity Fund and Toyota Ventures to create autonomous AI research labs.
Human AI researchers can't keep up with the pace of new AI research. They’re out of time to run the experiments they want. So we’re building an autonomous AI lab that can.
The era of human-scale R&D is over. Machine-scale development has begun. 🚀
Linda He retweeted
In middle school, with money earned performing magic shows I opened my first investment account.
At Stanford, captivated by the markets, I realized a paradigm shift was on the horizon.
Today, I am excited to announce our $5.7M raise, led by @novaholdings.
With the developments in AI, I knew I could no longer wait to start @vigillabs. At Vigil, we are building the engine to understand markets in realtime, designed for our own traders.
Linda He retweeted
introducing chipmunk—a training-free algorithm making ai video generation 3.7x & image gen 1.6x faster! ⚡️
our kernels for column-sparse attention are 9.3x faster than FlashAttention-3 and column-sparse GEMM is 2.5x faster vs. cuBLAS
a thread on the GPU kernel optimizations 🧵
Our latest joint work w/ SandyResearch @ UCSD: training-free acceleration of Diffusion Transformers w/ dynamic sparsity, led by @austinsilveria @SohamGovande!
⚡️ 3.7x faster video and 1.6x faster image generation while preserving quality!
🧵 Open-source code & CUDA kernels!
Excited to share our work on scaling LLMs to handle million-token contexts! Training models for ultra-long sequences is challenging due to data scarcity. We introduce a novel hierarchical synthetic data generation pipeline to overcome this. Thrilled this will be presented at ICLR 2025!
Paper: arxiv.org/pdf/2504.12637
Crucially, scaling up doesn't break performance on shorter contexts! Our 1M model maintains strong performance on standard benchmarks like LongBench (medium context tasks) and MMLU (general knowledge).
Takeaway: Our hierarchical synthetic data strategy provides a scalable and effective path to training LLMs for million-token contexts, enabling deeper understanding of long documents. Ablation studies confirm our hierarchical ordering is key! Work done @togethercompute
Check out the details & see you at #ICLR2025!
🔗arxiv.org/pdf/2504.12637
How does it work?
1️⃣ Start with long documents.
2️⃣ Create hierarchical summaries (global, medium, local).
3️⃣ Use an LLM (like Qwen) to generate diverse QA pairs: multi-hop, complex reasoning, local detail extraction.
4️⃣ Combine data across multiple docs for cross-document understanding.
This creates rich, structured data mimicking complex real-world tasks!
Linda He retweeted
🧬 Meet Lyra, a new paradigm for accessible, powerful modeling of biological sequences. Lyra is a lightweight SSM achieving SOTA performance across DNA, RNA, and protein tasks—yet up to 120,000x smaller than foundation models (ESM, Evo). Bonus: you can train it on your Mac.
read our paper here: arxiv.org/abs/2503.16351
Linda He retweeted
I taught an LLM to optimize proteins. It proposed a better carbon capture enzyme.
Introducing Pro-1, an 8b param reasoning model trained using GRPO towards a physics based reward function for protein stability.
It takes in a protein sequence + text description + previous experimental results, reasons over the information given in natural language, and proposes modifications to improve the stability of the given sequence.
🧵(1/n)
Linda He retweeted
Introducing Carl, the first AI system to create a research paper that passes peer review. Carl's work was just accepted at an @ICLR_conf workshop on the Tiny Papers track. Carl forms new research hypotheses, tests them & writes up results. Learn more: autoscience.ai/blog/meet-car…
Linda He retweeted
Read about how our RAG fine-tuning code assistant (fine-tuned Mistral-7B) can outperform both Claude 3 Opus and GPT-4o on some popular open source codebases!
We get up to 19% more accuracy, 3.7x faster speed, and 150x cost reduction.
together.ai/blog/rag-fine-tu…
