isildun

@isildun@sh.itjust.works

• ⁨Comment⁩ on ⁨How Quora Died⁩ ⁨⁨9⁩ ⁨months⁩ ago⁩:

  I’m almost starting to wonder if that’s the plan. Just keep saying “IPO IPO IPO” to get funding from over-eager VCs who want a piece of the IPO before it becomes widely available.

  But then you just never IPO. Keep making minor to moderate mistakes along the way so you can be all “weeeeell we would have IPO’d but insert thing here so we want to wait another 6 months to let it die down”. Repeat until you’re ready to quit, then actually IPO and ride the initial IPO high all the way down via golden parachute.

• ⁨Comment⁩ on ⁨OpenAI brings Sam Altman back as CEO less than a week after he was fired by board⁩ ⁨⁨11⁩ ⁨months⁩ ago⁩:

  Copilot, yes. You can find some reasonable alternatives out there but I don’t know if I would use the word “great”.

  GPT-4… not really. Unless you’ve got serious technical knowledge, serious hardware, and lots of time to experiment you’re not going to find anything even remotely close to GPT-4. Probably the best the “average” person can do is run quantized Llama-2 on an M1 (or better) Macbook making use of the unified memory. Lack of GPU VRAM makes running even the “basic” models a challenge. And, for the record, this will still perform substantially worse than GPT-4.

  If you’re willing to pony up, you can get some hardware on the usual cloud providers but it will not be cheap and it will still require some serious effort since you’re basically going to have to fine-tune your own LLM to get anywhere in the same ballpark as GPT-4.

• ⁨Comment⁩ on ⁨Results of the "Can you tell which images are AI generated?" survey⁩ ⁨⁨1⁩ ⁨year⁩ ago⁩:

  Image generation tech has gone crazy over the past year and a half or so. At the speed it’s improving I wouldn’t rule out the possibility.

  Here’s a paper from this year discussing text generation within images (it’s very possible these methods aren’t SOTA anymore – that’s how fast this field is moving): …thecvf.com/…/Rodriguez_OCR-VQGAN_Taming_Text-Wit…