Sign In

Josh Levitan

Technology Leader With Focus on Product & Scaling

Josh Levitan is a versatile product professional with a diverse background in marketing and operations. He has garnered significant experience in startup environments by spearheading teams, and enhancing product development processes. Josh holds a BFA in Film/Television from New York University. His extensive career includes pivotal roles such as Director of Product & Engineering at CommunityConnect Labs, Chief Product Officer at Breathwrk Inc., and Vice President Of Product Management at mPulse Mobile, Inc. Noteworthy positions also include VP, Product & Marketing at AIR.TV, Consultant at Levitan Consulting, CEO at Acquisition Sciences, VP Product & Marketing at LuckyLabs Inc., Founder, President at Gamzee, Inc., Senior Producer at Nexon, Senior Producer/Lead Game Designer at Playdom/Walt Disney Interactive Media Group, SEO/Web Consultant when self-employed, Project Manager at World Poker Tour, and Creative Executive at Goldmann Pictures/Clarion/Atmosphere Entertainment.

Highlights

Apr 2 · linkedin

This is insane.

Here's a Github repo for a ChatGPT-powered podcast generator (in alpha). This deserves all the stars.

https://lnkd.in/gTGd9Hqn

You plug in the topic of your choice and it autogenerates a roughly one-hour-long podcast with multiple sections on the subject of your choice, picks a voice, uses text-to-speech to narrate (it automatically picks out a voice based on the topic), and then stitches the whole thing together in a single mp3 file.

Anyone with an OpenAI API key and a few dollars can create decent enough long-form audio content.

Here are some examples I created with yesterday. You can see from the billing screenshot, this cost $4.42 to create 3 hours-plus of podcasts.

  1. David Bowie in his later years

https://lnkd.in/gsasc3F6

  1. The Boer War and its effects on Africa

https://lnkd.in/gyu29xR9

  1. The evolution of the Republican party in US politics over the last 25 years

https://lnkd.in/gBsiNGnB the evolution of the republican party in us politics over the last 25 years.mp3

Mar 21 · linkedin

More LLM goodness. I understand why models get things wrong (either they weren't trained on it, their training data is stale, or they made it up).

I don't understand why they forget basic capabilities they have or why they do weird things. Is this a function of older training data winning out (e.g. pre-GPT-4 models not being able to do search)?

ChatGPT (on the left) forgot it can search (again). When reminded, it can then search.

Gemini/Bard made up some extra instructions for itself, which is odd, because I asked it to confirm it understood my instructions and it did.

Josh Levitan
Add to my network