HNHacker News
TopNewBestAskShowJobs

mwigdahl

1,303 karma · joined July 18, 2017

submissionscomments
mwigdahl··on Service members deserve the right to repair
The military could not vote either. You had to _complete_ Federal Service before you earned the franchise.
mwigdahl··on Neuralink 'Participant 1' says his life has changed
My daughter has DBS for severe Tourette's. Her quality of life before the implant was horrible -- frequent 110 dB+ screams and self-injury. The implant has reduced her tic frequency and intensity by easily 95%. It's not only given her her life back but also the lives of her family members.

The potential of brain interface technology is truly incredible -- both for good and ill.

mwigdahl··on 1981 Sony Trinitron KV-3000R: The Most Luxurious Trinitron [video]
I don't, but I feel like GP was just expressing respect for the aesthetics of the engineering accomplishment, orthogonal to whether it was obsolete or not.

Something akin to how one might feel looking at the design of a clipper ship.

mwigdahl··on What is going on right now?
Perfect! You really got to the core of the matter! The only thing I noticed is that your use of the em-dash needs to not be bracketed with spaces on either end. LLMs—as recommended by most common style guides—stick to the integrated style that treats the em-dash as part of the surrounding words.
mwigdahl··on Tiny microbe challenges the definition of cellular life
That is not dead which can eternal lie...
mwigdahl··on OpenMower – An open source lawn mower
Still hasn't happened to me; beeping smoke alarms drive me crazy.
mwigdahl··on Perplexity offers to buy Google Chrome for $34.5B
Funding secured!
mwigdahl··on Claude Code is all you need
You can run Claude Code in the terminal window of VS Code, and it has IDE integration so you can see diffs inline, etc. It's not fully integrated like Cursor but you get a lot of the benefits of the IDE in that way.
mwigdahl··on Claude Code is all you need
Interesting that the other vendors haven't done this "trivial" task, then, and have pretty much ceded the field to Claude Code. _Every_ CLI interface I've used from another vendor has been markedly inferior to Claude Code, and that includes Codex CLI using GPT-5.
mwigdahl··on GPT-5: Key characteristics, pricing and system card
It seems like few people are referencing the improvements in reliability and deception. If the benchmarks given generalize, what OpenAI has in GPT-5 is a cheap, powerful, _reliable_ model -- the perfect engine to generate high quality synthetic data to punch through the training data bottleneck.

I'd expect that at some level of reliability this could lead to a self-improvement cycle, similar to how a powerful enough model (the Claude 4 models in Claude Code) enables iteratively converging on a solution to a problem even if it can't one-shot it.

No idea if we're at that point yet, but it seems a natural use for a model with these characteristics.

mwigdahl··on Vibechart
It's been fixed on the OpenAI website.
mwigdahl··on GPT-5 for Developers
Has anyone tried connecting up GPT-5 to Claude Code using the model environment variables?
mwigdahl··on GPT-5: Key characteristics, pricing and system card
If reasoning is on the table, then you already had to add o3-mini-high, o3-mini-medium, o3-mini-low, o4-mini-high, o4-mini-medium, and o4-mini-low to the 4.1 variants. The GPT-5 way seems simpler to me.
mwigdahl··on Claude Opus 4.1
<raises hand> Our automated test folks were chronically behind, struggling to keep up with feature development. I got the two assigned to the team that was the most behind set up with Claude Code. Six weeks later they are fully caught up, expanding coverage, and integrating AI code review into our build pipeline.

It's not 10x, but those guys do seem like they've hit somewhere around 2x improvement overall.

mwigdahl··on Claude Opus 4.1
Yep: https://docs.anthropic.com/en/docs/claude-code/sub-agents
mwigdahl··on AI is not Making Engineers 10x as Productive
I've heard this multiple times (Tea being an example of problems with vibe coding) but my understanding was that the Tea app issues well predated vibe coding.

I have experimented with vibe coding. With Claude Code I could produce a useful and usable small React/TS application, but it was hard to maintain and extend beyond a fairly low level of complexity. I totally agree that vibe coding (at the moment) is producing a lot of slop code, I just don't think Tea is an example of it from what I understand.

mwigdahl··on Fine-tuned small LLMs can beat large ones with programmatic data curation
Thanks for the explanation and the clarification on terminology! I've used a similar approach myself and it sounded like you were doing something similar.
mwigdahl··on Fine-tuned small LLMs can beat large ones with programmatic data curation
Is this just distillation but with a step to filter out low-quality responses first?
mwigdahl··on Sleep all comes down to the mitochondria
Proof? Doesn't need to be specific -- a general study showing higher all-cause mortality in Ozempic users compared to a control group over a long period would be just fine.
mwigdahl··on Study mode
If someone has Grok 4 access I'd be interested to see if it's less likely to avoid these specific issues.
mwigdahl··on Stop selling “unlimited”, when you mean “until we change our minds”
To quote the movie Zoolander: "I feel like I'm taking crazy pills!"

Anthropic never sold Max plans as unlimited. There are two tiers, explicitly labeled "5x" and "20x", both referring to the increased usage over what you get with Pro. Did all the people complaining that Anthropic reneged on their "promise" of unlimited usage not read anything about what they were signing up to pay $100 or $200/month for? Or are they not even customers?

mwigdahl··on Stop selling “unlimited”, when you mean “until we change our minds”
That would be a great argument if they ever promised unlimited usage. They didn't.
mwigdahl··on Stop selling “unlimited”, when you mean “until we change our minds”
I was going to write much the same thing but just upvoted you instead. Max plans are a super great deal, even with these new limits.
mwigdahl··on Stop selling “unlimited”, when you mean “until we change our minds”
It was never sold as unlimited. Max plans have always had both rate limits and 5-hour usage limits.
mwigdahl··on Using Claude Code Off-Label: Research Instead of Coding
This is an interesting application, makes me wonder how much of the "secret sauce" in Claude Code is truly coding-oriented and how much is just superior agentic logic in general.
mwigdahl··on How Anthropic teams use Claude Code
I’ve seen it firsthand at work, where my developers are shy about spending even a single digit number of dollars on Claude Code, even when it saves them 10 times that much in opportunity cost. It’s got to be some kind of psychological loss aversion effect.
mwigdahl··on AI overviews cause massive drop in search clicks
There are tons of companies selling bottled air. Here's a story from 7 years ago. There are lots more now: https://www.theguardian.com/global/2018/jan/21/fresh-air-for...
mwigdahl··on Hyperpb: Faster dynamic Protobuf parsing
Really missed a great naming opportunity with "superpb" (pronounced as "superb").
mwigdahl··on AccountingBench: Evaluating LLMs on real long-horizon business tasks
Larger in magnitude or in count of digits?
mwigdahl··on I avoid using LLMs as a publisher and writer
Sure, compilers are deterministic and LLMs are not. If you're asserting that a probabilistic process can't get you to a deterministic outcome, Monte Carlo integration would like to have a word.

My point was that comparing the rise of AI tooling to the rise of HLL compilers is a much better comparison than comparing it to crypto.

HLL compilers were originally seen as crutches and inferior tools and that "real" programmers used assembly. Compiler-generated code was derided as inefficient and ugly.

And it was! In the early days, a good programmer who knew the machine could outdo the compiler. But that didn't stop a huge expansion of new programmers who could write COBOL and FORTRAN but never learned assembly. And the compilers got better over time. These days it's a rare wizard who can outdo a compiler's optimizations, and it takes multiple orders of magnitude longer for those rare humans to achieve it.

LLM tooling isn't going away. Even in these very early days, it enables non-programmers to construct basic applications that work, using English requests! And the tools have gotten better on almost a monthly basis.

You can like them or not like them, just like the early programmers could like or not like compilers. But dismissing them as analogous to empty crypto hype is a bad comparison.

← PreviousPage 8 of 16Next →