josh halliday
75 posts
josh halliday
@LLMenjoyerUK
Building code, Frontier evals, benchmarking, GDPval and RL datasets at Turing, I also strap cams to collect robotics data. 20+ yrs as a music producer.
Brighton, England Katılım Ağustos 2024
710 Takip Edilen81 Takipçiler
@yuris @LuelCompanyAI i was thinking what the market needs is another human data company
English
Super excited to colead @LuelCompanyAI’s $31.2M seed round.
There are certain teams you meet where you know within 5 minutes that you want to partner with them Luel is one of those team.
William and Inigo are incredibly ambitious founders who understand the human data bottleneck from the inside out. They've built Luel to create a scalable, reliable supply of that data — something that will be foundational to the next generation of AI.
English
A key takeaway from GDPval:
It’s often less about the "arbitrary domain" (the topic) and more about the underlying model behavior it reinforces.
The real move is using unit tests to audit your verifiers.
It stops the model from rewarding hallucinated logic and forces it to prioritize code that actually executes.
If it doesn't pass the test, the reward shouldn't exist. 👍
English
Data Collector uses our @gi_labs UMI grippers to do welding at a construction site ( video credit: @GeorgiZlatarev )
English
josh halliday retweetledi
run your own research lab over a weekend!
poolside@poolsideai
setup hell kills good RL ideas. so we’re giving researchers Laguna XS.2, @PrimeIntellect Lab, and a weekend in London to run the whole loop: tasks → evals → rewards → training → rollouts → adapters → inference 14 days to go. come touch the weights: luma.com/poolsidehackat…
English
In case anyone is wondering how much:
- $15,000 for the codebase
- $30,000 for the company workspace
Yeah, nobody values your shitty code.
The Information@theinformation
AI Agenda: Why Turing Is Buying Up Failed Startups’ Codebases Struggling startups are considering selling their codebases to AI labs. Read more from @Steph_Palazzolo 👇 thein.fo/3N6pSLJ
English
@catboosted anonymous anime cat profile turns out to be a petulant child, go figure
English
@catboosted and in your little mind you think we just bend over and pay up whatever they ask for? you don’t think we have quality criteria?
-emailed a single broker
-decided you know everything
-proceed to die on a hill
English
@LLMenjoyerUK I’m sure you pay more because startups don’t give a fuck when dumping assets but brokers can charge whatever markup they want
English
@catboosted well done for emailing a broker i’m sure that gave you highly accurate feedback on the market landscape
English
@catboosted the startups are all charging wildly different amounts for their data, so again, no you don’t know what you are talking about.
Source: I actually work in this industry
English
@LLMenjoyerUK Idk what the labs are buying it for but I do know what the startups are receiving
Source: block me and never come back if you don’t like my tokens, I’m not asking you to believe me
English
@SeanZCai blinding interview, loved the insight to cyber RL tasks 🔥
English
Some alpha here!
Chris Barber@chrisbarber
Yesterday I interviewed @SeanZCai about AI data. This is essentially a guide for founders on how to sell data and RL envs to AI labs. "I've never seen a data contract get turned down by a top lab, if it's good quality data, for budget reasons." 00:00 What areas of data are underserved? 02:10 For bio data, is it real-world or purely digital? 04:21 For cyber data, which subsets are most underserved? 05:50 What is the sales process like? 07:04 Why would a lab not renew or increase their purchase volume? 10:13 When a researcher is exploring a new direction, what's the first step? 11:35 In robotics data, what do you view as underserved? 13:12 What does the initial data delivery look like, what format? 13:53 Do labs have more sophisticated internal setups for running environments? 14:32 Are the non-frontier labs buying off-the-shelf data from Anthropic / OpenAI vendors? 16:11 Do Anthropic data vendors put expiry timeframes on the exclusivity? 16:42 Are purchase decisions researcher-led? 17:41 Decagon, Sierra, Ramp: what kinds of data are they buying? 19:06 Long-term, when do labs still need to buy external data vs train on user traces? 21:15 Will end-vendor benchmarks shift to performance per dollar? 22:04 How many labs are spending at the 1B+/yr data level? 23:53 Delta between Anthropic's stated $1B and your 10-20B/lab number? 26:05 What makes inference providers / neoclouds a good fit to acquire RL env cos?
English
@davematthews @altantutar lol everyone’s buying ego 2D / 3D at the moment, human data collections onsite, mostly with go pros but moving to higher fidelity stereo cams now
English
@altantutar Can you talk more about the ego-centric data deals they can’t talk about? Why so secretive?
English
Robotics founder starter pack (save it):
→ SLAM jokes worse than your dad
→ "who buys robodogs anyway?"
→ "actually, teleop is not that bad"
→ best time in 20 years to build this
→ ego-centric data deals you can't talk about
→ Shenzhen, Shenzhen, Shenzhen, and maybe SF on a good day
→ "is this a bubble?" asked at the VC dinner, not in the warehouse
→ a customer who wants the outcome, not the robot
→ "we actually use an LLM for the planner"
What else?
altan tutar@altantutar
I talked to 10+ robotics operators, customers, and investors in the last week. Here's what I learned: 1/ Roboticists are building robots for themselves, not for the customers. 2/ Physical AI is harder than LLMs. We need way more data, and how we collect data is more important ever. 3/ Humanoids are over-valued. We don't need general robots & human-looking legs for many tasks. People are underestimating the power of building robotics vertically for certain tasks. 4/ People are paying big bucks for ego-centric data. Foundational labs are cutting exclusive deals with data brokers. 5/ There is disagreement as to whether "robotics is a bubble." Some people believe that we overfunded robotics companies. Other people believe that we have not even scratched the surface. 6/ A lot of founders play the VC game (e.g. nice humanoid robot videos to get more funding), rather than understanding the needs of a customer. 7/ Teleop is undervalued and is doing way more work than people admit. 8/ China is eating the component & humanoid stack. 9/ Customers don't care about your robots. They want outcomes cheaper and efficient. 10/ The best robotics teams aren't just pure ML & AI PhDs. They have a weird mix of hybrid backgrouds. I'm curious what others are seeing on the ground. What am I missing?
English