Hacker News: ValentineC

New comment by ValentineC in "Discovery Loop"

ValentineC — Wed, 05 Aug 2026 16:33:16 +0000

I wonder if Jeff Dean facts [1] (I hope people remember the reference) will carry over to the new startup.

[1] https://github.com/LRitzdorf/TheJeffDeanFacts

New comment by ValentineC in "Helsinki Hacker News Meetup"

ValentineC — Wed, 05 Aug 2026 14:52:00 +0000

Anyone interested in a meetup in Singapore?

New comment by ValentineC in "Increasing the lifespan of a bulb makes it worse in every other way"

ValentineC — Fri, 31 Jul 2026 22:26:54 +0000

Randomly, my Philips Hue E27 bulbs last far longer than any of the China Zigbee clones I've bought, but like all mass produced goods, there are outliers that fail.

The cheap China stuff tend to fail in a few months, while I've only had two Hue bulbs in my collection fail ever, so Philips seems to be doing something right.

New comment by ValentineC in "Apple Will 'Watch Everything Burn' When AI Bubble Bursts"

ValentineC — Fri, 31 Jul 2026 22:10:47 +0000

> Microsoft too is a cash cow and seemingly hasn't bet the farm on Copilot.

I think a good follow-up question is: which Copilot?

Anyway, Microsoft seems to be leaning hard into the whole "you can't get fired for buying Microsoft" thing that used to be IBM and AWS. Normies all around the world still insist on using Microsoft Windows and Office.

Anecdotally, they supposedly lost a lot of money on GitHub Copilot's inference costs, at least until June. So they'd be far more equipped to survive an AI crash now.

New comment by ValentineC in "DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis"

ValentineC — Fri, 31 Jul 2026 16:32:51 +0000

@dang please update the existing 404 link to https://artificialanalysis.ai/models/deepseek-v4-flash

New comment by ValentineC in "The AI Aesthetic"

ValentineC — Fri, 31 Jul 2026 01:10:40 +0000

Beige/cream colours and orange accents — wait, isn't that literally the HN design?

(Randomly, I love the Solarized colour schemes, so anything similar is great, vibe-coded or not.)

New comment by ValentineC in "Kimi K3-256k"

ValentineC — Thu, 30 Jul 2026 21:41:24 +0000

You're right that it's model dependent, even within Claude models.

I've found Opus 5 far better as a subagent with very limited context window use, which could suggest that it might not have good long context performance unlike its predecessors. (I was one of many tearing my hair out trying to work with Opus 5 for the past week.)

New comment by ValentineC in "Launch HN: Tokenless (YC S26) – Automatic model switching to save money"

ValentineC — Thu, 30 Jul 2026 02:29:31 +0000

Claude's cache breaks with a reasoning effort adjustment:

https://platform.claude.com/docs/en/build-with-claude/effort...

New comment by ValentineC in "Ask HN: Is everybody's Claude weekly quota reset time the same?"

ValentineC — Thu, 30 Jul 2026 02:09:03 +0000

It's fixed per account, unlike ChatGPT's which starts when you use it.

Epic diarrhea outbreak has 40% of Americans avoiding fruits and veggies

ValentineC — Tue, 28 Jul 2026 23:29:24 +0000

Article URL: https://arstechnica.com/health/2026/07/epic-diarrhea-outbreak-has-40-of-americans-avoiding-fruits-and-veggies/

Comments URL: https://news.ycombinator.com/item?id=49091406

Points: 10

# Comments: 2

Opus 5: RL-Fried and mistake-prone for anyone else?

ValentineC — Tue, 28 Jul 2026 20:14:10 +0000

Article URL: https://old.reddit.com/r/ClaudeAI/comments/1v92csh/opus_5_extremely_rlfried_and_mistakeprone_for/

Comments URL: https://news.ycombinator.com/item?id=49089313

Points: 5

# Comments: 0

New comment by ValentineC in "Benchmarking Opus 5 on SlopCodeBench"

ValentineC — Tue, 28 Jul 2026 02:45:37 +0000

Medium or low supposedly prevents Opus 5 from overthinking:

https://xcancel.com/danshipper/status/2080700057892815114

New comment by ValentineC in "Elevated errors on Claude Opus 5"

ValentineC — Mon, 27 Jul 2026 22:47:02 +0000

I mentioned this in one of the earlier Opus 5 threads [1], and it's still infuriatingly bad after days.

I've read that Opus 5 has more success if used with Fable as orchestrator, but since I refuse to pay for subscription access to Fable, I'm now trying Opus 4.6 as orchestrator (also since that was the version that got me loving Claude), and Opus 5 low effort as implementor.

[1] https://news.ycombinator.com/item?id=49052980

Anyone else's human get nerfed this week?

ValentineC — Mon, 27 Jul 2026 15:08:05 +0000

Article URL: https://old.reddit.com/r/ClaudeAI/comments/1v7zlxd/anyone_elses_human_get_quietly_nerfed_this_week/

Comments URL: https://news.ycombinator.com/item?id=49070750

Points: 4

# Comments: 2

New comment by ValentineC in "What is happening to jobs? Separating AI hype from reality"

ValentineC — Sun, 26 Jul 2026 00:12:55 +0000

I'm surprised the layoffs worldwide hasn't produced more entrepreneurs building their own software product.

New comment by ValentineC in "The new rules of context engineering for Claude 5 generation models"

ValentineC — Sat, 25 Jul 2026 23:41:35 +0000

I haven't been impressed with Opus 5 over the past ~30 hours either.

It's made countless careless mistakes folding in plan amendments after they get reviewed by Sol, and has produced sloppy mockups (e.g. buttons overflowing past cards) despite all the supposed verification claims.

New comment by ValentineC in "The new rules of context engineering for Claude 5 generation models"

ValentineC — Sat, 25 Jul 2026 22:40:16 +0000

Commit Strip seems to be down, so here's the Wayback Machine version:

https://web.archive.org/web/20260521130338/https://www.commi...

New comment by ValentineC in "Claude Opus 5"

ValentineC — Sat, 25 Jul 2026 18:54:05 +0000

Ah, thanks for clarifying!

New comment by ValentineC in "Claude Opus 5"

ValentineC — Sat, 25 Jul 2026 16:25:03 +0000

What I think is more alarming is that Claude's advice for prompting Opus 5 says that Opus will verify its own work [1]:

> Claude Opus 5 verifies its own work without being told to. If your prompt contains explicit verification instructions ("include a final verification step for any non-trivial task," "use a subagent to verify"), remove them: instructions like these cause over-verification on Claude Opus 5, and removing them reduces wasted tokens with no loss in quality.

My own experience over the past ~20 hours hasn't been great either, with Opus producing sloppy mockups (e.g. buttons overflowing past cards) without doing any of the purported verification.

[1] https://platform.claude.com/docs/en/build-with-claude/prompt...

New comment by ValentineC in "Claude Opus 5"

ValentineC — Sat, 25 Jul 2026 03:36:28 +0000

I wonder if this is one of the few times simonw's pelican was broken on the first try [1]:

https://tools.simonwillison.net/markdown-svg-renderer#url=ht...

My experience with Opus 5 thus far haven't been that great either. It's been making mistake after mistake editing my coding plans that were being reviewed by GPT-6 Sol.

[1] https://simonwillison.net/2026/Jul/24/introducing-claude-opu...