Mihai M

AI Whisperer | Founder of

[Mihai M @mihaim]

@lucaspatiri_ UGC

[Mihai M @mihaim]

@davidfigeira Yapper

[Mihai M @mihaim]

@irabukht I’m looking forward to testing your platform. Does it do ad set / campaign analysis only or also generates AI ads? How much does it go into campaign management?

[Mihai M @mihaim]

@mckaywrigley Show me

[Mihai M @mihaim]

Reinventing Espresso Tonic on first principles

[NEXTA @nexta_tv]

Coffee with mineral water becomes a viral trend

The new drink — coffee with mineral water — has quickly gained popularity online, with claims that it boosts energy more effectively than any energy drink.

The secret is that the mineral water helps caffeine absorb faster, while also replenishing magnesium and calcium lost due to coffee.

The recipe is simple: a glass with ice, mineral water, a shot of espresso, a pinch of salt, and a bit of lime juice.

[Mihai M @mihaim]

We're making a longevity iOS app meant to help people expand their health span and generally live a longer, healthier life.

Just opened the private beta. I'd like to personally invite you to sign up! https://t.co/hOhKo3SPXl

[Mihai M @mihaim]

To this day, my favorite film ever

[John Carmack @ID_AA_Carmack]

Deep down at the bottom of Hephaestus’ giant forge, a charred arm sticks out of the glowing molten metal with its thumb held high.

[Mihai M @mihaim]

Vibe coding has a happy medium where you also have tests and read the code carefully before committing to git

I think the voice only fast iteration is mostly a meme that almost nobody is actually doing in a meaningful way

[Erwin @Erwin_AI]

Vibe coding is basically code for "vibe now, stress later" when you're confronted with angry customers, impossible bugs and code you don't understand even one bit.

[Mihai M @mihaim]

Bring back small phones please

[Tim Cook @tim_cook]

Introducing the newest iPad Air!

[François Chollet @fchollet]

The key is really this: AI usefulness scales logarithmically with inference time compute. Right now for many use cases the amount of compute you need to operate at human-level is such that AI isn't economically viable for that use case.

The more compute efficient AI gets, the more use cases start becoming economically viable, the more we'll deploy AI, and the more compute we'll need.

[Mihai M @mihaim]

@bryan_johnson @emigal

[𝐌𝐚𝐭𝐭 𝐏𝐢𝐧𝐧𝐞𝐫 @Matt_Pinner]

You may be old but, are you this old??

[Mihai M @mihaim]

This is the plot of 'Black Mirror' in ad form

[Tim Cook @tim_cook]

Meet the new iPad Pro: the thinnest product we’ve ever created, the most advanced display we’ve ever produced, with the incredible power of the M4 chip. Just imagine all the things it’ll be used to create.

[kepano @kepano]

@tim_cook I think the ad would work much better if it was reversed. All the objects should be expanding out of the iPad rather than being crushed into it

made this edited version in five minutes (thanks iMovie!)

[Mihai M @mihaim]

@paulg @vladciurca

[Patrick McKenzie @patio11]

@kendrictonn ... that it takes basically one incident of sufficient magnitude to decisively remind relevant network actors that it doesn't stop being a LARP the moment you decide it stops being a LARP, it stops being a LARP the moment Power decides it stopped being a LARP.

[Mihai M @mihaim]

@jxnlco Is response_model an undocumented parameter? Can't find it in the @OpenAIDevs API Reference. Can only find response_format, which can mean a json, but not the model adherance.

[Mihai M @mihaim]

@jonclemonsjp @jessethanley @nikitabier Say more

[Yann LeCun @ylecun]

I've made that point before:
- LLM: 1E13 tokens x 0.75 word/token x 2 bytes/token = 1E13 bytes.
- 4 year old child: 16k wake hours x 3600 s/hour x 1E6 optical nerve fibers x 2 eyes x 10 bytes/s = 1E15 bytes.

In 4 years, a child has seen 50 times more data than the biggest LLMs.

1E13 tokens is pretty much all the quality text publicly available on the Internet. It would take 170k years for a human to read (8 h/day, 250 word/minute).

Text is simply too low bandwidth and too scarce a modality to learn how the world works.

Video is more redundant, but redundancy is precisely what you need for Self-Supervised Learning to work well.

Incidentally, 16k hours of video is about 30 minutes of YouTube uploads.

[Simon Sarris @simonsarris]

Knowledge is almost never the bottleneck. Competent and engaged people are. They are still very, very rare.

MIT knows this, they know the value isn't in the open courses, its in the network. 99% of the value of college is making and maintain ties, not things you can learn online