Manuel Del Verme
@ManuelDelVerme
Transparent, interoperable FOSS AI is the great equalizer. I invest//advocate//build for developers accelerating progress toward expanded human potential 🚀🌟
This is a big step towards agents that are both general and reliable. Having general and realistic benchmarks is what allowed machine learning as a field to progress so quickly
🧵-1 We are thrilled to release #AgentLab, a new open-source package for developing and evaluating web agents. This builds on the new #BrowserGym package which supports 10 different benchmarks, including #WebArena.
1/7 For the past decade, our team at Meta Reality Labs (previously CTRL-labs) has been dedicated to developing a neuromotor interface. Our goal is to address the Human Computer Interaction challenge of providing effortless, intuitive, and efficient input to computers.
🎯Our agents now achieve a 94% top 5 success rate on ServiceNow tasks, up from the previous SOTA of 63.8%! That means an over 11-minute half-life, up from 1 minute. 📚How did we reach this milestone? Deep dive 👉 silverstream.ai/blog-news/2sig… 👇Watch our agent in action! #BPO…
Totally aligned with @kimberlytan’s “Unbundling BPO” thesis. After 500+ hrs of customer interviews at @SilverstreamAI we see the same: browser agents cut silos by acting like humans, no ad-hoc integrations needed. Buyers care first about ROI, then reliability & data ownership.
📢 New @a16z thesis: AI will unbundle the BPO and disrupt the $300b outsourcing market. Enterprises often outsource important, but high-volume and repetitive work to BPOs. These BPOs rely heavily on human labor, often leading to slow turnaround times, human error, and…
I am back in Montreal. Glad Silvestro (our AI agent) booked the trip. That saved me some time.
Special thanks to our swarm of web agents who worked hard to gather data for Pasta-1T ! Can't wait to show more of their work!
We published a deep dive into Pasta-1T, the largest dataset of real-world web trajectories for browser agents! 🌐 silverstream.ai/blog-news/past…
Excited to finally present Silverstream AI to the world. Agents should be accessible to everyone. Agents want to be free.
🚀 Meet @silverstreamAI. We're happy to announce that we raised a $1.2M pre-seed round to build the foundation for reliable Autonomous Web Agents. This round was led by the incredible teams at @GradientVC, with strong support from @vento_ventures. This financing round supports…
If you do the math, operator type agents should be priced free for everyone, 200$/mo is nowhere close to justified
”If you trained on the open web, your model should be open source.” perfect take by @naval // this is a decent way of dealing with AI’s intellectual property challenges.
LLaMA and Deepseek make me think about the rise of open source. What key moments might have hindered its benefits? Were there close calls that nearly stopped its growth? is there a 10 hours podcast on the "History of OSS"?
🌐 Internship Opportunity on web agents at ServiceNow Research! 🚀 We aim to boost 📈 the performance in #WorkArena and #WebArena by fine-tuning large open-source LLMs. (Candidate must be affiliated with a Canadian university) 🔗 Apply Here: bit.ly/3UQCk3X
How capable are web agents at solving knowledge work tasks? 🤔 Are LLMs up to the challenge? 🤖 Introducing WorkArena: a benchmark where agents meet the world 𝘸𝘪𝘭𝘥 web of enterprise software 🌐🖥️ Paper: bit.ly/4a7FiFV Website: bit.ly/3VkdJ87 🧵 1/7
Link to the full post, tutorial notebooks, and code is here: lesswrong.com/posts/kobJymvv…
I'm excited to release Prisma, a mechanistic interpretability library for multimodal models like CLIP and ViTs. Incubated at @tyrell_turing's lab & in collab with @NeelNanda5. Recent mech interp work has focused on language, but many techniques transfer. Behold, the dogit lens:
You want your personal AI to experience the world through your perspective. This way, as agents improve your AI will be able to do your bidding autonomously because it will totally understand you. But beyond sight and sound What other data should it absorb to fully grasp your…
Check out this new paper: Led by @mehdiazabou and @evadyer, we show that it is possible to get SOTA brain decoding with transfer across individuals and tasks! The key is a clever way to tokenize spiking data for transformers. #brain #neurotech #NeurIPS2023
Is a universal brain decoder possible? Can we train a decoding system that easily transfers to new individuals/tasks? Check out our #NeurIPS2023 paper where we show that it’s possible to transfer from a large pretrained model to achieve SOTA 🧠! Link: poyo-brain.github.io 🧵
Introducing the Animal-AI Olympics, a new kind of AI competition. Instead of providing a problem to solve, we provide an arena in which we will test for simple cognitive abilities using experiments inspired by the animal cognition literature. More details: animalaiolympics.com