Harvard AI Professor: The People Most at Risk From AI Aren’t Who You Think

Comments · 120

• SI

  @SiliconValleyGirl · 4 days ago · pinned

  ChatLLM by Abacus AI is the tool featured in this episode. It brings leading AI models together in one place: <a href="https://chatllm.abacus.ai/svg">https://chatllm.abacus.ai/svg</a>

  10

• IS

  @issiewizzie · 4 days ago (edited)

  I said this many years ago. Anyone can Google but not everyone is a professor

  17

• QU

  @quiet_one8061 · 4 days ago

  As some one who uses AI daily it feels like he is behind the curve when it comes to AI using antiquated techniques and what he has read instead of working with it daily.

  43

• NG

  @nguyễnhuy-x6s4d · 4 days ago

  When the marginal cost of expertise drops to zero via AI, the ultimate competitive edge shifts entirely to human taste, strategic imagination, and the courage to set uncompromising standards.

  10

• CO

  @Complex-AR · 1 day ago (edited)

  <a href="https://www.youtube.com/watch?v=vFOnDsy58NU&amp;t=309">5:09</a><br>A <b>Verifier</b> can be built into the LLM, so that <b>verification</b> and <b>assurance</b> can be applied to each response to grade the veracity of any claims or statements of fact, e.g. verified, partly verified, contested or unverified.<br>Also, a <b>Writing Standard</b> should be applied to remove <b>mannered prose</b> from the LLM&apos;s response, as it can cause meaning and context to drift or be lost over long reasoning sessions, especially after compaction.<br>There should be a <b>Consistency Checker</b> applied to each response to catch any drift or wrong information present in the LLM session.<br>All of these defences can be applied in several layers or places in an LLM to work simultaneously on each response, e.g. in the global instructions or as skills or as plugins.

  2

• FR

  @FragVantaHQ · 4 days ago

  This is literally me trying to figure out which of the 50 AI tools I opened today is the right one. 🙃

  2

• US

  @userexperienceresearchers · 5 hours ago

  The distinction between using AI to do more and using it to learn more is becoming increasingly important. What resonated with me is that expertise, taste, and judgment do not develop by simply accepting the first answer. We still need to understand the fundamentals, challenge the machine, and take responsibility for the outcome. AI can expand what we are capable of, but only if we keep growing alongside it.

• FE

  @ferb9990 · 4 days ago (edited)

  <a href="https://www.youtube.com/watch?v=vFOnDsy58NU&amp;t=750">12:30</a><br><br>Models are trained not only to predict statistically mean token. It is only the first step of training.<br><br>There is also RL part, where models forced to solve tasks and such training makes them generate right solutions but not most likely next token

  6

• JI

  @jiroug · 2 days ago

  The premise of the professor is that AI is a tool just like one of the other tools people use in their daily life, maybe a little more advanced tool. But the goal of AI companies is to provide businesses with not a tool rather a replacement of the individual employee, which is in line with what CEO-s are looking for (replace as many individuals as possible with AI agents and tools). So, what should be the next generation preparing for? Use AI as a tool or recognize the fact that Agents will be taking over sooner or later? Sooner seems much more realistic - unfortunately.

  1

• CR

  @craigH2805 · 3 days ago

  Got a laugh out of the boss who did not do his own email. My boss was like that. Someone asked if he had a computer and he answered yes. But I pointed out it was sitting in a box under his desk.

  3

• AL

  @AlstonAnalytics · 3 days ago

  I think I do the common sense thing. I usually start a project by having three different frontier models ask questions about the project for deep research. I then synthesize those questions and de deep research on that list. And then write out planning and architecture documents. Etc. Don’t people usually start with a prompt that goes, “Make sure all recommendations are evidence based, consider my trade-offs, aligning with my ultimate goals…”? I think the mindset shift is moving from browser UI usage to creating systems in an IDE or solid harness. Good luck all.

• SA

  @sajithkumarmohandasan5352 · 3 days ago

  AI has changed what we call smart. Memorising isn’t the key anymore. Thinking clearly, asking better questions and using information well is what matters now. The world isn’t less smart. It’s just looking for a different kind of intelligence.

  1