HNHacker News
TopNewBestAskShowJobs

mwigdahl

1,303 karma · joined July 18, 2017

submissionscomments
mwigdahl··on Why jet engines aren't made in China
Hence the fading away of the "nerd with pocket protector" stereotype...
mwigdahl··on Fable 5 is Back
During the initial release, they indicated you'd be able to use Fable 5 as part of your subscription for a limited period of time, and then it would require usage credits. They also did say at that time that they hoped to make it a part of the subscription plans again at some point after that.

For this return, they've extended the usage period to July 7, but limited you to 50% of your usage quota, and have not restated the desire to make it permanently part of the subscription plans at some point.

I still have hope, but it's not moving in the right direction to be sure.

mwigdahl··on Beware, Claude Code deletes >30 day old transcripts. Anthropic won't fix it
Sure, and if they hadn't trimmed logs someone would also be complaining that "stupid Anthropic lets logs fill your hard drive"!

Oh wait, they just complained about Codex doing this a day or so ago.

mwigdahl··on Claude Sonnet 5
It's a good question, but for multiturn conversations even cached context adds up quickly. My experience has been that spawning off subagents for defined tasks in a large overall plan generally makes me come out ahead.

I'm sure folks' mileage will vary though.

mwigdahl··on Claude Sonnet 5
I don't think all the negativity is from crazies, but big chunks of it are certainly motivated. I certainly left out numerous other categories.
mwigdahl··on Claude Sonnet 5
For agentic computer use Sonnet 5 low performs better than Sonnet 4.6 medium at just under half the cost, and better than Opus 4.8 low at 25% off. Their success rates are not that far off.

Agentic search is a different story, but even there it still dominates 4.6 (as in, for everything Sonnet 4.6 can do, Sonnet 5 can do it as well or better at the same or lower cost).

Yes, Opus 4.8 dominates Sonnet 5 over its entire range in both categories, but Opus's lower range is limited and there is a valid regime on the lower end where Sonnet 5 use makes economic sense. This is not the case for Sonnet 4.6 where Opus 4.8 dominates it completely on both charts.

Edit -- reading your response closer I think we're saying the same things, maybe just disagreeing on whether that lower end is valuable or not.

mwigdahl··on Claude Sonnet 5
Seems like a combination of multiple factors:

"They took my shit away!" -- 3-day Fable 5 addicts (me)

"How dare they tell Trump no?" -- US nationalist / "my country right or wrong" types

"Great to see a closed source company fail!" -- open source boosters

"Great to see an American company fail!" -- anti-US, and/or pro-China folks

"Great to see a successful company fail!" -- anti-capitalists and/or sour-grapes crab bucket types

"Serves you right for ripping off creators!" -- copyright warriors

"They keep silently nerfing the models!" -- secret downgrade conspiracy theorists

"Quit killing the planet!" -- anti-datacenter advocates

mwigdahl··on Claude Sonnet 5
What? If you're comparing their models in the same size class, Sonnet 5 is Pareto-optimal over Sonnet 4.6.
mwigdahl··on Claude Sonnet 5
Not to sound like an LLM, but that seems exactly right to me. Use it as a cheaper, high-functioning task subagent and lower reasoning for a master Opus session. As long as not every portion of your task requires maximum intelligence, you should come out ahead.
mwigdahl··on County with 37 Data Centers Asks Schools to 'Conserve Electricity'
Was that tech guy a crypto miner by chance?
mwigdahl··on Beware, Claude Code deletes >30 day old transcripts. Anthropic won't fix it
"Evil Anthropic implements rolling log files."

Film at 11.

mwigdahl··on Beware, Claude Code deletes >30 day old transcripts. Anthropic won't fix it
Except that those customers can access the traces for 30 days, and freely copy them at any point during that period? It's a usability issue at worst, not some anti-customer conspiracy.
mwigdahl··on I'm building a Space Cadet Pinball Machine! [video]
Wow, that's fascinating! What else can you tell us about the boat? How big was it? What was the rigging? How long did they take on the circumnavigation?

I've always dreamed of doing something like that.

mwigdahl··on Age verification is just a precursor to automated attribution of speech
Jury time is paid time off at my US company. And at least at the US Federal level there's a daily stipend for sitting on one. Lower level courts may vary on that.
mwigdahl··on Elevated error rate across multiple models
I'll do you one better! Why is GLM 5.2?
mwigdahl··on The AI Ceiling Is Lower Than Anyone Is Saying
Article is clearly slop.
mwigdahl··on A robot is sprinting towards you. Do you want it running on Claude or Grok?
"It's the smell, if there is such a thing. I feel saturated by it. I can taste your stink and every time I do, I fear that I've somehow been infected by it."

Agent Smith, _The Matrix_

mwigdahl··on Bitcoin and energy: the no-bullshit version, with sourced numbers
Nine rhetorical uses of the word "honest" (and one used as descriptor for the network).

Sloppy writing.

mwigdahl··on Sixty percent of US consumers say 'AI' in brand messaging is a turnoff
You wouldn't need AI for this; deterministic programming would be enough (and scads cheaper).
mwigdahl··on Workers are spending over 6 hours a week botsitting AI, fueling job frustration
Same here. The power upgrade going to Fable in particular is quite impressive.
mwigdahl··on Age verification tech could put children at greater risk, says think tank
Using the built-in Screen Time tools, yes. Qustodio works pretty well though as an add-on product. Not perfect UX, not perfect functionality, but it's the best I've found.
mwigdahl··on C++: The Documentary
Agreed, _Modern C++ Design_ is probably the programming book I've gotten the most out of in my career.
mwigdahl··on I built a vulnerable app and spent $1,500 seeing if LLMs could hack it
An easy way around the API token thing is to put it in a file and point the model at the file. I saw what you were seeing when I provided credentials directly, but haven't had any problems with it since using the indirect method.
mwigdahl··on Expertise in the age of AI
This effect is certainly real, likely even the default, but it's not inevitable.

I had great results using AI to help my son study for his final exams in chemistry and math. We went through the review guide the teacher provided, he did the problems, I checked them, and I had Claude generate additional targeted problems as permutations of the ones he had difficulty with. He worked them and got more practice in exactly the areas he was weak.

I could have set these problems up myself, but it was much smoother to have Claude set them up and I validate them. It let him get a lot more reps in, in exactly the areas he _needed_ more practice, than he would have otherwise.

The key is that to learn you have to do the work. AI can help you figure out where you're weak and provide the wherewithal to get additional practice, and there's huge value there. But you have to lift the mental weights yourself.

mwigdahl··on The last six months in LLMs in five minutes
“Am I out of touch?”

“No, it’s the children who are wrong.”

mwigdahl··on Click (2016)
Is it capturing and selling user data? Avast would certainly know.
mwigdahl··on Ask HN: What's your go-to LLM for coding?
Opus 4.7 in Claude Code. I've used GPT-5.5 in Codex and it's good also, but I have a pretty developed workflow in Claude Code so it's hard to tell if GPT-5.5 is actually rougher or if it just doesn't have the same tooling support I've built up in CC over the past 6 months.

I should add that it's not perfect -- nothing in this space is -- but it to me has the best level of both consistency and quality of the setups I've tried.

mwigdahl··on Cursor Introduces Composer 2.5
One of those conditionals is a pretty huge assumption.
mwigdahl··on The people writing AI alignment policy are not whose work is being replaced
Teilhard de Chardin was, for one.
mwigdahl··on The limits of Rust, or why you should probably not follow Amazon and Cloudflare
vi would be much easier to rewrite in Rust.
← PreviousPage 3 of 16Next →