Ayush Mandowara Haven't tried audio calls yet. I use ffmpeg to record video calls from my laptop. Nothing on the phone yet though. Do let me know your experiences once you get it working

Rohan Sen wow! What was the prompt?

Vijay Rangarajan (VJ) Interesting! How often do you run your AMAs? How do you socialize? How do you filter? How do you share the discussion? Any tips on the preparation or process? Who else do you know who does this well? Lots of questions - would love to learn from your experience!

I'm off for a 10-day Vipassana meditation program.

𝗪𝗛𝗔𝗧? A 10-day residential meditation. No phone, laptop, or speaking. https://www.dhamma.org/
𝗪𝗛𝗘𝗡? From today until next Sunday (13 July)
𝗪𝗛𝗘𝗥𝗘? Near Chennai. https://lnkd.in/geGVFynY
𝗪𝗛𝗬? I've heard good things and am curious.
𝗦𝗨𝗥𝗘? I've never been away from tech for this long. Let's see!

I've scheduled LinkedIn posts, so you'll still see stuff. But I won't be replying.

If someone asked me, "What's changed this year in LLMs", here's my list:

Prompt engineering is out. Evals are in. https://lnkd.in/gfF-6giV
Hallucinations are fewer and solvable by double-checking. https://lnkd.in/gUhPp22n
LLMs are great for throwaway code / tools. https://lnkd.in/g8u4jsYH
LLMs can analyze data. No more Excel. https://lnkd.in/gdMBtW6y
LLMs are good psychologists. https://lnkd.in/gAJuFnsh
Image generation is much better. https://lnkd.in/g4Tg_cDB
LLMs can speak well enough to co-host a panel. https://lnkd.in/gQ59apG4
... and create podcasts. https://lnkd.in/gvCAdDpm

But:

LLMs are still not great at slides. https://lnkd.in/gQQwG_a6
LLMs still can't follow a data visualization style guide.
LLMs can't yet create good sketch notes.
LLMs still draw bounding boxes as well as specialized models.
Agents (LLMs running tools in a loop) can think only for ~6 min.

What's on your list of things LLMs still can't do?

LLMs are smarter than us in many areas. How do we control them?

It's not a new problem.

𝗩𝗖 𝗽𝗮𝗿𝘁𝗻𝗲𝗿𝘀 evaluate deep-tech startups.
𝗦𝗰𝗶𝗲𝗻𝗰𝗲 𝗲𝗱𝗶𝘁𝗼𝗿𝘀 review Nobel laureates.
𝗠𝗮𝗻𝗮𝗴𝗲𝗿𝘀 manage specialist teams.
𝗝𝘂𝗱𝗴𝗲𝘀 evaluate expert testimony.
𝗖𝗼𝗮𝗰𝗵𝗲𝘀 train Olympic athletes.

… and they manage and evaluate "smarter" outputs in 𝘮𝘢𝘯𝘺 ways:

𝗩𝗲𝗿𝗶𝗳𝘆. Check against an "answer sheet".
𝗖𝗵𝗲𝗰𝗸𝗹𝗶𝘀𝘁. Evaluate against pre-defined criteria.
𝗦𝗮𝗺𝗽𝗹𝗶𝗻𝗴. Randomly review a subset.
𝗚𝗮𝘁𝗶𝗻𝗴. Accept low-risk work. Evaluate critical ones.
𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸. Compare against others.
𝗥𝗲𝗱-𝘁𝗲𝗮𝗺. Probe to expose hidden flaws.
𝗗𝗼𝘂𝗯𝗹𝗲-𝗯𝗹𝗶𝗻𝗱 𝗿𝗲𝘃𝗶𝗲𝘄. Mask identity to curb bias.
𝗥𝗲𝗽𝗿𝗼𝗱𝘂𝗰𝗲. Re-running gives the same output?
𝗖𝗼𝗻𝘀𝗲𝗻𝘀𝘂𝘀. Ask many. Wisdom of crowds.
𝗢𝘂𝘁𝗰𝗼𝗺𝗲. Did it work in the real world?

For example, you can apply them to:

𝗩𝗶𝗯𝗲 𝗰𝗼𝗱𝗶𝗻𝗴: Non-programmers might glance at lint checks (𝘊𝘩𝘦𝘤𝘬𝘭𝘪𝘴𝘵) and see if it works (𝘖𝘶𝘵𝘤𝘰𝘮𝘦).
𝗟𝗟𝗠 𝗶𝗺𝗮𝗴𝗲 𝗱𝗲𝘀𝗶𝗴𝗻𝘀: Developers might check if a few images look good (𝘚𝘢𝘮𝘱𝘭𝘪𝘯𝘨) and check a few marketers (𝘊𝘰𝘯𝘴𝘦𝘯𝘴𝘶𝘴).
𝗟𝗟𝗠 𝗻𝗲𝘄𝘀 𝗮𝗿𝘁𝗶𝗰𝗹𝗲𝘀: An journalist might run a 𝘊𝘩𝘦𝘤𝘬𝘭𝘪𝘴𝘵, a 𝘋𝘰𝘶𝘣𝘭𝘦-𝘣𝘭𝘪𝘯𝘥 𝘳𝘦𝘷𝘪𝘦𝘸 with experts, and 𝘝𝘦𝘳𝘪𝘧𝘺 critical facts (𝘎𝘢𝘵𝘪𝘯𝘨).

You 𝘢𝘭𝘳𝘦𝘢𝘥𝘺 know many of these. You learnt them in Auditing. Statistics. Law. System controls. Policy analysis. Quality engineering. Clinical epidemiology. Investigative journalism. Design critique.

Worth brushing up on these skills. They're 𝘮𝘰𝘳𝘦 important in the AI era.

ChatGPT: https://lnkd.in/g-q6jttw

I catch up on long WhatsApp group discussions as podcasts.

The quick way is to scroll on WhatsApp Web, select all, paste into NotebookLM, and create the podcast.

Mine is a bit more complicated. Here's an example:

• Use a bookmarklet to scrape the messages https://lnkd.in/gXXHQK28
• Generate a 2-person script https://lnkd.in/gQqEYWpM
• Have 𝚐𝚙𝚝-𝟰𝚘-𝚖𝚒𝚗𝚒-𝚝𝚝𝚜 convert each line using a different voice https://www.openai.fm/
• Combine using 𝚏𝚏𝚖𝚙𝚎𝚐 https://ffmpeg.org/
• Publish on GitHub Releases https://lnkd.in/gEUiCbpZ

I run this every week. So far, it's proved quite enlightening.

Podcast: https://lnkd.in/gc7BDN6E: https://lnkd.in/gcqgWWVh