Hon Weng Chong's Cortical Labs sells CL1 devices that keep living human and mouse neurons alive, with over 3,000 cloud sign-ups for a service that serves only 20 or 30 people at once.
Turing CEO Jonathan Siddarth argues that private evaluations tailored to a lab's own query distribution matter more than public benchmarks, which he says are academic, synthetic and increasingly saturated.
Mercor CEO Brendan Foody says his marketplace matches AI labs with specialized contractors averaging over $90 an hour, and that the top 10% of workers drive most project value.
Cortical Labs faces demand far beyond its neuron computer supply
Funding and deals
Cortical Labs. Builds computers from living neurons. A new round, amount not given, funds CL1 delivery, with Blackbird Ventures and Horizons Ventures among backers.
Guest interviews
Hon Weng Chong, CEO and cofounder of Cortical Labs. His team taught neurons to play Pong with structured electrical signals instead of rewards or shocks, and its CL1 device keeps neurons alive in a sealed environment, sometimes for months. He says the original experiment used about 800,000 to a million neurons and far less power than a 20 watt human brain. Demand is far beyond supply, with over 3,000 cloud sign-ups for a system that serves only 20 or 30 people, and he finds human neurons outperform mouse neurons.
Jonathan Siddarth, founder and CEO of Turing. He says public benchmarks reveal model gaps but are often academic and synthetic, so private evals built on a lab's own queries matter more. SWE-bench rose from 2% to 50% last year, now tops 60% and will likely saturate this year, so Turing is building a harder coding benchmark where all models start near zero. He says Turing employs PhDs to break models and write RL data, and is at nine figures of revenue, profitable and growing.
Brendan Foody, cofounder and CEO of Mercor. He wants one global job marketplace that pairs the reach of LinkedIn and Indeed with the value capture of recruiters, who typically take about 30% of first-year pay. Mercor now matches specialized contract talent to AI companies at an average pay above $90 an hour. He reports 1605% annual net revenue retention and profitability throughout, with annual recurring revenue reported around $75m.
News
Bettors see GPT-5 arriving by mid-August. Polymarket gives about 25% odds of release by August 5 and roughly two thirds by August 15. Alex Wilhelm says Sam Altman's hype sounds like past launches, but the improvement is real.
Rival labs flood the market with new models. xAI's Grok 4, Moonshot's Kimi K2 and Z.ai's GLM 4.5 arrived while Alibaba's Qwen models keep improving. OpenAI is also expected to release an open weights model soon.
Meta's Scale AI deal opens room for rival evaluators. Meta's investment in Scale AI and its absorption of Scale's CEO and cofounder created an opening for other evaluation companies, Alex Wilhelm says.
Tap a name to go there · tap a line to see why they are linked