Hacker News: vitorgrs

New comment by vitorgrs in "GLM-5.3: Frontier coding with emergent cyber capabilities"

vitorgrs — Fri, 14 Aug 2026 15:13:24 +0000

Just a warning, this is on Opus. There's not a clear harness winner. It will change depending on the models.

New comment by vitorgrs in "DeepSeek Harness developer preview"

vitorgrs — Thu, 13 Aug 2026 16:46:39 +0000

I just started testing, but didn't figured out if it already support MCP/plugins? It seems it can already use Deepseek search if it uses official API, but what about custom providers? Can we use together with MCPs like tavily?

New comment by vitorgrs in "Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users"

vitorgrs — Fri, 07 Aug 2026 19:05:27 +0000

That's also using AI :).

ChatGPT already competes with Google...

New comment by vitorgrs in "Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users"

vitorgrs — Fri, 07 Aug 2026 05:27:00 +0000

Luna is not bad, but... you feel is a small model. It have less knowledge. It's ok for purely agentic stuff (web searching etc).

New comment by vitorgrs in "Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users"

vitorgrs — Fri, 07 Aug 2026 05:23:45 +0000

If you just want to know "When it's the next full moon", yes. Very bad as google can answer in 1s.

New comment by vitorgrs in "Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users"

vitorgrs — Fri, 07 Aug 2026 05:19:15 +0000

There was already thinking in free model though. They removed the option a while ago and made it automatic... But you could force by just saying "think hard" or whatever.

New comment by vitorgrs in "Qwen3.8-Max: A New Bar for Coding and Cowork"

vitorgrs — Mon, 03 Aug 2026 05:57:04 +0000

OpenAI can't even build Codex app (not cli) for Linux... I thought code was solved?!

Ironically it's Electron BS, so actually, Sol could probably build itself...

New comment by vitorgrs in "Seedance 2.5"

vitorgrs — Sun, 02 Aug 2026 06:53:51 +0000

There is a ton of vertical dramas now (Duanju) made purely of AI. Even on chinese streaming services you find it.

New comment by vitorgrs in "Flux 3"

vitorgrs — Fri, 24 Jul 2026 08:16:15 +0000

Yeah, very weird launch. I was like... where's the videos?

New comment by vitorgrs in "Passkeys were invented by engineers with zero understanding of consumer brain"

vitorgrs — Thu, 23 Jul 2026 05:32:29 +0000

That might change on the country. At least here in Brazil, all logins require some sorth of MFA.

This changes by banks, some send code to our phone number (thought WhatsApp or SMS), others send SMS+email + face ID. All of them require at least the face ID. Some biggest banks requires you to go to ATM to authorize app access. You insert your card, password and authorize there.

There's Mercado Pago, which supports passkeys and standard MFA too. So you can store on bitwarden even.

OpenBench – A benchmark for comparing coding-agent harnesses

vitorgrs — Wed, 22 Jul 2026 00:57:25 +0000

Article URL: https://twitter.com/mattlam_/status/2079605387121049605

Comments URL: https://news.ycombinator.com/item?id=49000462

Points: 5

# Comments: 0

New comment by vitorgrs in "Oh-my-pi: A coding agent with the IDE wired in"

vitorgrs — Wed, 22 Jul 2026 00:53:11 +0000

I really like it, especially how it can easily get search from Codex/Gemini/etc.

New comment by vitorgrs in "Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber"

vitorgrs — Tue, 21 Jul 2026 23:13:41 +0000

Yes? And even for simple interactions, they use SVG by default.

New comment by vitorgrs in "Qwen 3.8"

vitorgrs — Mon, 20 Jul 2026 22:25:40 +0000

One of my tests to see if "we are there", it's to ask a list of mayors from my city. Sounds silly, but most models just create random mayors that never existed.

Without search, of course. In theory this should be easy, because clearly the Wikipedia data it's in the model.

New comment by vitorgrs in "Qwen 3.8"

vitorgrs — Mon, 20 Jul 2026 00:57:13 +0000

And I remember Sam Altman saying 2 years ago or more, that hallucination was already "fixed" internally. Well, clearly not.

New comment by vitorgrs in "Qwen 3.8"

vitorgrs — Sun, 19 Jul 2026 11:52:38 +0000

SVG's pelican https://gist.github.com/vitordelucca/521c2d63c9b852c622e7648...

Made on the website, so not sure if on the API there's more thinking options...

New comment by vitorgrs in "Qwen3.8 is launching and going open-weight soon"

vitorgrs — Sun, 19 Jul 2026 10:17:10 +0000

Xi Jinping openly talked about open source at WAIC. So don't think the labs have much a choice now...

New comment by vitorgrs in "Qwen 3.8"

vitorgrs — Sun, 19 Jul 2026 10:13:45 +0000

Deepseek 4 "final" version is imminent as well.

Will probably be at Opus 4.8 level, and I find it pretty big deal because of Deepseek price...

New comment by vitorgrs in "NYC may require landlords and realtors to disclose the use of AI in listings"

vitorgrs — Sun, 19 Jul 2026 05:22:52 +0000

I wish there would be a similar thing for food as well. In my country (Brazil), food apps such as iFood are FULL of AI images. Worse is that iFood don't allow reviews to post pictures, so you get very blind asking it.

Now Didi came (99) and Keeta, and they at least allow users to post reviews with pictures.

New comment by vitorgrs in "Kimi K3: Open Frontier Intelligence"

vitorgrs — Fri, 17 Jul 2026 08:17:02 +0000

phone number is the main way China services uses (I think it's even mandatory for social networks, as phone number there function almost like "ID")