Skip to content
OpenAIOpenAI

Episode 16: Building AI for Life Sciences

What does it take to build AI systems that can actually help scientists? Research lead Joy Jiao and product lead Yunyun Wang discuss how OpenAI is developing models for life sciences and what responsible deployment means in a field with real biosecurity stakes. They explore how AI is already improving research workflows and where it could lead in drug discovery and more autonomous labs — including why a future with less pipetting sounds pretty good to most scientists. Chapters 0:39 Introducing the Life Sciences model series 3:47 Joy’s path into life sciences 5:00 Autonomous lab with Ginkgo Bioworks 7:27 Yunyun’s path into life sciences 8:12 OpenAI’s life sciences work 9:48 Biorisk, access, and safeguards 15:43 What models can do in the lab 17:51 Building scientific infrastructure 20:14 Why compute matters for science 24:54 Where are we in 6-12 months? 29:51 Scientific adoption and skepticism 33:17 Advice for students and researchers 40:27 Where are we in 10 years?

Joy JiaoguestYunyun Wangguest
Apr 15, 202644mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

OpenAI’s Life Sciences models: accelerating labs, discovery, and safety safeguards

  1. OpenAI is launching a Life Sciences model series focused on biochemistry workflows, starting with genomics and protein understanding for early discovery use cases.
  2. The team describes “model-in-the-loop” autonomous lab work (e.g., with Ginkgo Bioworks) showing models can design experiments that yield measurable results and speed iteration cycles.
  3. Life-sciences deployment requires orchestration, reproducibility, and workflow templates (e.g., a plugin with 50+ repeatable skills) across products like ChatGPT and Codex.
  4. Because biology is highly dual-use and intent is hard to infer from prompts, OpenAI emphasizes differentiated access, enterprise controls, and incremental deployment to manage biorisk.
  5. They argue that scaling compute—both larger models and test-time “thinking time”—is central to enabling harder scientific reasoning, with a long-term vision of autonomous, robot-run labs guided by humans.

IDEAS WORTH REMEMBERING

5 ideas

Life-sciences AI must be built around real research workflows, not just chat.

They frame the Life Sciences model series as anchored on complex “long trajectory” research tasks—literature synthesis, pathway analysis, target selection—and delivered through product surfaces (ChatGPT, Codex) plus workflow templates for repeatability.

Tool use already makes models behave like competent computational biologists.

Joy highlights that models can call established tools (e.g., protein structure predictors), inspect outputs, and iteratively tweak inputs—mirroring how computational biologists run analyses and refine hypotheses.

Autonomous labs shift the bottleneck from human hands to compute.

The Ginkgo collaboration is used to illustrate a future where parallelized agent workflows and robotic execution reduce “human throughput” constraints (pipetting, protocol translation), making compute and orchestration the limiting factors.

Biosecurity is hard because benign and harmful steps look similar.

They emphasize that prompts like “help me clone a gene” are ambiguous (GFP vs. toxin), making intent detection unreliable; this motivates cautious defaults (refusals or high-level guidance) for general access.

Differentiated access is positioned as the key to unlocking stronger capabilities safely.

For verified researchers and regulated institutions (tracked reagents, controlled datasets, enterprise security), they argue more capable assistance can be provided than for anonymous general users, where misuse risk is harder to bound.

WORDS WORTH SAVING

5 quotes

We’re really excited to build and deploy the Life Sciences model series.

— Yunyun Wang

One of the taglines was to scale test time compute to cure all disease.

— Joy Jiao

The precursor steps… look very benign, and it’s really hard to distinguish between.

— Yunyun Wang

The safest model here would be a model that just had no capability… and it’s not very good, but it’s very safe.

— Joy Jiao

Nothing in biology is really real until you can prove it in the real world.

— Joy Jiao

Life Sciences model series (biochemistry-focused)Genomics and protein mechanistic understandingAutonomous wet lab collaboration (Ginkgo Bioworks)Model orchestration, agents, and reproducible workflowsLife sciences research plugin (50+ skills)Biorisk, information hazards, and differentiated accessCompute scaling and test-time compute for scientific discovery

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome