Hacker News: hmate9

New comment by hmate9 in "Gemini 3.6 Flash"

hmate9 — Tue, 21 Jul 2026 15:24:38 +0000

This model is not for builders and engineers. DeepSWE score of 49% is behind gpt 5.4 and muse spark. It's clearly intended to be an efficient model for google gemini usage.

What is interesting is how this is announced before any Gemini Pro progress. From the outside it seems as though Google cannot keep up with other frontier models.

New comment by hmate9 in "Show HN: One More Letter"

hmate9 — Thu, 16 Jul 2026 07:59:27 +0000

Thank you guys for the feedback. I'll fix valid words being rejected, having space as a keyboard shortcut for shuffle, and maybe a "zen" mode without a timer.

Show HN: One More Letter

hmate9 — Wed, 15 Jul 2026 23:10:10 +0000

Article URL: https://playonemoreletter.com/

Comments URL: https://news.ycombinator.com/item?id=48928402

Points: 91

# Comments: 59

New comment by hmate9 in "Previewing GPT‑5.6 Sol: a next-generation model"

hmate9 — Fri, 26 Jun 2026 17:15:06 +0000

What is this prediction based on?

New comment by hmate9 in "Micron locks in historically high memory prices for five years"

hmate9 — Fri, 26 Jun 2026 08:56:20 +0000

You can pay the most if you can get the most value out of it

New comment by hmate9 in "Identity verification on Claude"

hmate9 — Sun, 21 Jun 2026 21:11:50 +0000

Not defending this, and it's far from ideal, but also credit card details were already pretty much confirming identities already.

New comment by hmate9 in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

hmate9 — Sat, 13 Jun 2026 06:40:31 +0000

This is incredibly bullish for china and open source models

New comment by hmate9 in "DiffusionGemma: 4x Faster Text Generation"

hmate9 — Wed, 10 Jun 2026 17:05:11 +0000

I can’t help but feel like there’s something here that will matter for future LLMs.

The bidirectionality could be a big deal: being able to refine a sentence with both left and right context feels closer to how editing/thinking actually works than committing to each token forever.

Maybe the current models aren’t good enough yet, but the direction feels important.

New comment by hmate9 in "Gemini 3.5 Flash"

hmate9 — Tue, 19 May 2026 20:49:33 +0000

I have google ai pro plan and tried antigravity with 3.5 flash but it used up all my quota in two prompts. If that is not a bug then it is seriously unusable.

Musk vs. OpenAI Verdict: Musk Lost

hmate9 — Mon, 18 May 2026 18:07:47 +0000

Article URL: https://twitter.com/ns123abc/status/2056436278682669112

Comments URL: https://news.ycombinator.com/item?id=48183239

Points: 5

# Comments: 1

New comment by hmate9 in "Updates to GitHub Copilot interaction data usage policy"

hmate9 — Wed, 25 Mar 2026 20:39:17 +0000

going to github i was greeted with a banner and a link directly to the settings for changing it

New comment by hmate9 in "Updates to GitHub Copilot interaction data usage policy"

hmate9 — Wed, 25 Mar 2026 19:46:31 +0000

For what it's worth they're not trying to hide this change at all and are very upfront about it and made it quite simple to opt out.

OpenAI Parameter Golf

hmate9 — Wed, 18 Mar 2026 17:17:14 +0000

Article URL: https://openai.com/index/parameter-golf/

Comments URL: https://news.ycombinator.com/item?id=47428478

Points: 12

# Comments: 2

New comment by hmate9 in "Meta and TikTok let harmful content rise to drove engagement, say whistleblowers"

hmate9 — Tue, 17 Mar 2026 20:55:05 +0000

I think the burden to curate your feed so that you do not have such content is now resting with the user and they cannot rely on the platform to do it for them.

New comment by hmate9 in "Meta and TikTok let harmful content rise to drove engagement, say whistleblowers"

hmate9 — Tue, 17 Mar 2026 20:54:11 +0000

Is this unavoidable? I mean it does generate clicks and views and user engagement so if one platform is doing it, doesn't that automatically mean that the other has to do it? Otherwise they will continuously lose market share.

Ask HN: Are past LLM models getting dumber?

hmate9 — Tue, 10 Feb 2026 12:06:57 +0000

I’m curious whether others have observed this or if it’s just perception or confirmation bias on my part. I’ve seen discussion on X suggesting that older models (e.g., Claude 4.5) appear to degrade over time — possibly due to increased quantization, throttling, or other inference-cost optimizations after newer models are released. Is there any concrete evidence of this happening, or technical analysis that supports or disproves it? Or are we mostly seeing subjective evaluation without controlled benchmarks?

Comments URL: https://news.ycombinator.com/item?id=46958617

Points: 4

# Comments: 5

New comment by hmate9 in "Testing Ads in ChatGPT"

hmate9 — Mon, 09 Feb 2026 19:29:46 +0000

I hate ads too but whats the outrage? Did people expect it to be free forever? Everything else has ads. youtube, instagram, x, google etc. etc.

New comment by hmate9 in "xAI joins SpaceX"

hmate9 — Mon, 02 Feb 2026 21:58:36 +0000

Any details regarding valuations etc?

New comment by hmate9 in "The Codex App"

hmate9 — Mon, 02 Feb 2026 19:50:32 +0000

That’s not really a negative for me as I can easily jump into vscode where I already have my workspace for coding set up exactly as I like it. This being a completely separate app just to get the agentic work right is a good direction imo

New comment by hmate9 in "Kimi Released Kimi K2.5, Open-Source Visual SOTA-Agentic Model"

hmate9 — Tue, 27 Jan 2026 10:05:01 +0000

About 600GB needed for weights alone, so on AWS you need an p5.48xlarge (8× H100) which costs $55/hour.