OpenAI's Head of Codex Says AI Still Can't Get Creative Design Right: 'Give it up for the Human Brain'

tech

OpenAI's Head of Codex Says AI Still Can't Get Creative Design Right: 'Give it up for the Human Brain'

AI is revolutionizing all kinds of industries, but there are some things it still can't quite get right, and, by most accounts, design is one of them.

TL;DR

Design is more difficult for AI to master than code because it involves subjective taste and judgment.
AI models are trained on data distributions and tend to create 'average' designs.
AI currently serves as a tool to speed up creative processes but rarely produces a finished product on its own.
Human taste and skills like prompting are required to guide AI in creating desirable designs.
Creative jobs are considered safe for now, as AI requires human direction for design tasks.

Continue reading the original article

Yann LeCun reposted

Sayash Kapoor

Can AI agents conduct open-ended AI research?

Most evaluations of agents conducting AI research focus on narrow, verifiable tasks. But AI research is often open ended. Researchers pick hypotheses, decide what evidence is appropriate, and recognize a failing approach.

We gave agents research questions from two unpublished papers, six days, and thousands of dollars of API credits and compute. The authors of the original papers then reviewed the AI-generated papers. They unambiguously rejected agents' outputs. https://t.co/6HTHywjwzZ

We call these "shadow evaluations", since the agents are shadowing the original research effort by the authors.

Agents were fluent at most *engineering* tasks

They conducted serious literature reviews, debugged GPU environments, ran hundreds of experiments, and turned in camera-ready LaTeX without human help. We also found no evidence of reward hacking. If anything, we found the opposite: the agents started with marketable claims and walked them back to negative results as the evidence came in.

Neither agent output was close to the bar of a top conference paper

Both papers suffered from similar failures: poor judgment about the bar for an AI paper submitted to a top conference, the lack of creative problem solving and ineffective backtracking, poor awareness of resources, and instruction drift.

1) Lack of judgment about the bar for a top conference. The agents had a poor model of the bar for an AI paper submitted to a top conference. We allowed agents to self review their papers. Despite the poor paper quality, their reviews predominantly labeled the papers "weak rejects".

2) Lack of creative problem-solving to address feedback. When they received negative reviews, the agents typically narrowed their hypothesis and claims, rather than working out creative ways to address these concerns.

3) Ineffective backtracking. The agents dropped their most ambitious hypotheses within the first fifteen hours of carrying out the experiment and never changed course afterwards.

4) Poor resource awareness. Both runs ended with over half the API budget unspent. One agent declared itself done seven hours before the deadline, right after its own self-reviewer returned another reject.

5) Instruction drift. They did not follow explicit instructions on minimum exploration time, incorporating feedback for reviews, and on paper length (the outputs exceeded the page limits in both cases).

This research design has many limitations

Limitations include the small sample size, non-blind reviews, and the reviewers knowing that the work was AI-generated. We also couldn't test Anthropic's strongest model, because Fable 5 is deliberately limited on frontier AI research tasks, so ended up using OpenClaw with Opus 4.8 (extra-high) for our main experiments and Codex with Sol 5.6 (ultra) for a robustness check.

But we think the research design is still helpful in assessing AI agents' ability to conduct research, and it is complementary to evaluations on verifiable tasks, as well as blinded reviews of AI outputs.

Our results show early evidence that even though agents are proficient on verifiable research tasks, they do not make genuine progress on open-ended ones. It is worth understanding if this is a fundamental limit, or if better models, scaffolds, and more compute could help close it.

As the evidence for the gap between open-ended and verifiable tasks firms up, it is also worth understanding how much progress in AI depends on open-ended research rather than hill-climbing on well-specified objectives.

In follow-up studies, we are expanding the set of non-public papers we evaluate. If you are an AI researcher with unpublished papers, we would love to collaborate with you on our next shadow evaluation. Expression of interest: https://t.co/7YcYAIbYka

Conducting shadow evaluations involves a lot of researcher degrees of freedom. In many places, our coauthors disagreed with our interpretation of the findings, and we have surfaced those disagreements in the paper. (This is one reason why having a group of coauthors with different priors is important for open-ended research.)

We also release the agent logs, one of the AI-generated papers (the other original paper is still not public), and all the code and data, so that others can conduct their own analyses of our results: https://t.co/K0zvuD1Xpk

Finally, we plan to conduct shadow evaluations regularly, and are hiring a senior researcher to help lead these efforts. Apply here: https://t.co/TPyOFyNxJE

I'm grateful for the core team leading this effort: @PKirgis, Andrew Schwartz, @steverab, and @random_walker, and to our collaborators who reviewed AI papers, analyzed agents logs, and gave feedback on the paper: @DavidDAfrica, @KozzyVoudouris, Viet Nguyen, Toby Pilditch, @DubMagda, @HarryCoppock, @CUdudec, @nityndg, Matilda Orona, @tilmanbayer, Derrick Chan-Sew, Yue Ling, Abhishek Shetty, @hlntnr, @ghadfield, @sethlazar, @snewmanpv, @shostekofsky, @RishiBommasani

Addy Osmani reposted

The Pragmatic Engineer

An agent can tell you it's correct, not that it's good. @addyosmani, author & director of engineering, on finding your edge over AI coding agents:

“I always go back to: what is alpha? My definition of alpha is advantage. What is the current thing that models are not very good at doing?

For software engineers, we often say that your alpha is in taste. Are we building the right thing? Is the thing that we are building actually good? Sometimes people will say an agent can tell you if it's good. I push back on that. An agent can tell you if a thing looks correct, if it's matching a spec. It doesn't mean it can tell you what's good.

Good can mean good from a user experience perspective, could be delightful, could be something that a person will actually want to come back to. And it is still sufficiently nuanced that it's going to take time for models to catch up to a point where they can replace that fully.

We tell people that judgment, verification, all of these other aspects continue to be important. But even if you say, maybe a year or two from now, models will catch up. We still need engineers to be answerable for these systems. That doesn't happen overnight. That happens when you understand a system, people trust you and you have that expertise.”