Piligrim

my Russian telegram-channel about AI/Tech news -

[Piligrim @PiIigr1m]

@brianchew @koltregaskes Hey, thanks for reply, I just got home to check. It's just make Codex work without creating Project, but I want to make it work in some specific folder (ex C:/cool_app/) with full access to it, but that it won't create Project.
Am I missing something obvious?

[Piligrim @PiIigr1m]

@btibor91 @koltregaskes Yes, I also remember something similar. What's new here?

[Piligrim @PiIigr1m]

@PauseAI Smart people will accelerate further

[Piligrim @PiIigr1m]

@himanshustwts It's just a joke and not real AI. They so behind even Chinese model that pretty much no one use their (or Yandex) models. *Much* popular choice here is DeepSeek/Perplexity, who don't want to use VPN.

Don't know about Sber, but Yandex, as told, just fine-tune Chinese models.

[Piligrim @PiIigr1m]

@test_tm7873 @varrock I read somewhere that Yandex is just fine-tuning Chinese models. I believe the one who is actually making models themselves is Sber, but I'm not sure. Anyway, I know no one who really uses Russian AI; it's mostly DeepSeek/Perplexity for those who don't want to use a VPN.

[Piligrim @PiIigr1m]

@imjustnewatai I wish, I sometimes really feel lack of limits for my Plus account...

[Piligrim @PiIigr1m]

@imjustnewatai Will we, someday, see a report, or something like this, that companies gave stronger models in full private to government? Or it will be just a rumor that never will be officially verified?

And when OAI will release a beast😭 I want RSI so bad

[Piligrim @PiIigr1m]

@imjustnewatai It's always government😁

[Piligrim @PiIigr1m]

@imjustnewatai Yoooo, back again? For how long?

[Piligrim @PiIigr1m]

@AILeaksAndNews Congratulations! Now we will have leaks and news from "master" himself 😀

[Piligrim @PiIigr1m]

@ChrissGPT Didn't thay have at least 900m WAU (weekly). So I don't think that 1b MAU is unexpected. I honestly thought that it's 1b WAU, but I guess we need to wait a little more

[Piligrim @PiIigr1m]

@HarshithLucky3 June 4? Or nah?

[Piligrim @PiIigr1m]

@testingcatalog Haiku 4.5 is used only to translate audio to text and to generate text-answer, speech is generated by ElevenLabs, right?

[Piligrim @PiIigr1m]

@Figure_robot I hope that in the end there will be statistics about performance and general work of robots (individually or combined) with some observations that was during this "shifts"

[Piligrim @PiIigr1m]

@scaling01 He always have same response from this 2 options:
1 - if LLM bad, say "I always told you that LLM is a scam"
2 - of LLM is good, say "This only because it use NeUrOsYmBoLiC tools, and I always told you that it's the future"

He just can't stop wining

[Piligrim @PiIigr1m]

@intology I want to see how they will perform with internet access. Sure, they can see the "best" solution, but i doubt that humans who set new records don't use previous attempts/information at all. And basic internet access can help with correct implementation of latest optimizations.

[Piligrim @PiIigr1m]

@btibor91 Isn't it's already possible to send Emails through Writing Block?

[Piligrim @PiIigr1m]

@DmitryRybin1 Maybe not on track for super-exponential, but a bit faster than exponential.

And METR said that they can't guarantee that these results are reliable and accurate - we just don't have tasks to measure that long time horizon.