HNHacker News
TopNewBestAskShowJobs

jameshart

22,635 karma · joined March 31, 2011

submissionscomments
jameshart··on The worst spam emails: iLands AI agent hustle
There’s some great thinking on this topic from Nicklas Lundblad in this Hannah Fry video on AI Agents: https://www.youtube.com/watch?v=WnzR5aOElvw

He talks about how agents/delegates extend the agency of the people who operate them. And “almost all of society is premised on the fact that agency is scarce”. But software agents grant vastly more agency.

The thing that stops the worst people from annoying the hell out of you is that they lack the agency to do so. But now…

jameshart··on A Design Space Exploration of Async/Await
To be fair to yourself, the fact that C# terminates pending tasks when the main method exits is something most C# devs don’t have to deal with because they’re mostly working in the context of long running servers and apps.
jameshart··on A misalignment of AI in mathematics
> destroyed the yardstick … that has traditionally been used to measure how much they have contributed

This goes much broader than mathematics or academia. This is the entire basis via which society distributes its wealth: based on a labour market derived valuation of ‘contribution’.

jameshart··on Kenyans Did College Students' Homework for Years. Then A.I. Arrived
You’re portraying a lot of these things as if they exist in a continuum along a fuzzy line from ‘not cheating’ to ‘cheating’, but the reality is it isn’t fuzzy at all. It’s cheating to submit work you didn’t do for assessment as if it is your own. That is cut and dried cheating.
jameshart··on Houthis 'take control' of key island in global shipping route
It’s not so much ‘we are not in a position to verify’ as the fact that the person who said the thing, and what they said, is itself newsworthy.

If you’re reporting that “Government sources say Houthis have ‘taken control’ of an island” then the fact that government sources used the words ‘taken control’ is itself interesting.

The headlines “Russia ‘will be defeated’ in Ukraine says Zelensky”, “Russia ‘will be defeated’ in Ukraine says Putin”, and “Russia ‘will be defeated’ in Ukraine says Taylor Swift” all tell of vastly different news events.

jameshart··on The Multiplayer AI Manifesto
Of course the best way to have your ideas dismissed is to label them a ‘manifesto’, but there’s some interesting ideas in here.

I’m particularly struck by the fear that we might be becoming less collaborative as developers turn away from one another and into chats with AI assistants - the idea of bringing those assistants out into the collaborative environment is compelling. Invite the agent to the meeting.

jameshart··on The Waymo effect: how AI is quietly making research less collaborative
This touches on some of the topics that drove Sergey Karayev to write about ‘multiplayer AI’ - https://fullstackdeeplearning.com/working-with-ai-agents/mul...

The fact that work is retreating into private chat sessions with AI agents is a loss; the idea of bringing those agentic collaborators out into public spaces to collaborate with teams is interesting.

jameshart··on Don't let anyone take away your big box of cables
We have the same. Do you know why the ‘every audio connector known to man’ box contain every audio connector known to man but yet never a stereo 1/8” headphone-jack to headphone-jack cable?
jameshart··on Don't let anyone take away your big box of cables
Put a USB C to A adapter on each end and you’ve got yourself a perfectly serviceable USBC cable.
jameshart··on Douglas Hofstadter: Analogy as the Core of Cognition [video]
Embedding spaces are all about analogy.

I’m confused why there seems to be a dismissal of the most basic ‘strange loop’ of the LLM - the fact that it’s evaluating a context to choose the next word, then reevaluating in a context where that word has been appended.

That always seemed to me like the essence of a Hofstadterish strange loop, so the emergence of Hofstadterish phenomena (self rep, etc) doesn’t seem surprising.

jameshart··on The UN challenges five centuries of cartography
With computer generated maps, you can just always use a minimally distorting conical or polar projection centered on the exact point you’re focusing on.

That also happens to be broadly what you would get looking straight down at the earth from above that point, which seems like it would also be the least confusing option.

jameshart··on The UN challenges five centuries of cartography
The prime meridian being the least skewed area of the equal earth and Robinson projections, with their bulgy nature, actually serves to make them both offer somewhat inconvenient depictions of places like Alaska and New Zealand and Japan.

Mercator is far more globally evenhanded in that regard, being a cylindrical projection.

jameshart··on Claude, change the “Add to Cart” button to blue
To test this, change the history in the context to indicate that the model did or recommended something completely different than it actually did, and then ask it to explain why. You’ll still get a plausible explanation.
jameshart··on Claude, change the “Add to Cart” button to blue
I learned this from Dirk Gently’s Holistic Detective Agency, which I strongly recommend.
jameshart··on Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses
Possibly an inevitability, since LLM text is specifically generated to be low entropy. It’s smooth, optimized for minimal perplexity.

And if you don’t find any of the word choices a little bit perplexing, your brain can just gloss over it. There is no meaning here. It’s like smalltalk.

jameshart··on How Climate Resilient Are the Largest Cities?
Chicago is lowest for adjusted risk because it just scores very low for physical risk on their scale. Compared to other cities the adjustment to its risk score for resilience is about typical - not great and not terrible - which suggests that yes, for what physical risks it does face (summer heat) it could probably do more to reduce those risks.
jameshart··on DaVinci Resolve 21.1
Joe said it right in Halt and Catch Fire:

> Computers aren't the thing. They're the thing that gets us to the thing.

That was that character’s take on Jobs’s ‘bicycle for the mind’ concept.

jameshart··on DaVinci Resolve 21.1
A lot of professional skill nowadays is misdirected from ‘knowing how to do the thing’ into ‘learning how to use a computer to do the thing’.

The role of music producer includes a large chunk of ‘knowing how to program synths and samples and drums and effects’. The role of industrial designer includes a lot of ‘knowing how to model things in CAD’. Even roles like academic researchers require skills like ‘formatting LaTeX’.

I’m genuinely quite excited for a shift in human computer interaction which reduces that ‘knowing how to do things on the computer’ tax.

But there are a whole swathe of jobs where a lot of what people do is ‘know how to do the thing on a computer’. Some of that is really because the bureaucracy has been embodied in a computer system so for example the job of handling an insurance claim seems to involve a lot of ‘knowing how to operate the claims computer system’ - actually that person’s skill is ‘knowing how to navigate the claims bureaucracy’.

jameshart··on Bill Gates tries to install MovieMaker (2003)
Probably phoned Ballmer, and Ballmer told him who to send his rant to.

The degree of overthinking going on in this thread…

jameshart··on Bill Gates tries to install MovieMaker (2003)
Lower level than the former CEO And President of the board.

The point is this was not a round robin reply all buckpassing exercise with BillG and the senior corporate leaders still on the reply list. This is a specific downstream chain of email.

I’m not sure I understand what people here think the right thing to do is when the company founder emails a bunch of senior execs telling them they seem to have let things all go to crap.

jameshart··on Bill Gates tries to install MovieMaker (2003)
Most of the thread is lower level staff specifically trying to address the raised issues. We have no idea what conversations this triggered among the exec team about the broader strategic issues Bill raised.
jameshart··on Bill Gates tries to install MovieMaker (2003)
This was an email Bill Gates sent in 2003. Not sure you can hold the ‘TikTok generation’ accountable for that.
jameshart··on Bill Gates tries to install MovieMaker (2003)
Really? You’ve got this guy Dave Fester on there saying he was going to own the website issues and was coordinating everything to do with how the products surfaced on the web. That seemed like pretty clear accountability and ownership to me. What do you want? Some more performative mea culpas?
jameshart··on Bill Gates tries to install MovieMaker (2003)
This email was literally BillG sending an email to the people he holds accountable for this, pointing out to them that he believes his experience resulted from a systematic lack of attention to end to end user experience.

I’m not sure why you see this as evidence BillG didn’t understand that to be his job.

jameshart··on What is Nueralese and Why is it Bad
Or the neuralese could be hidden steganographically in the chain of thought.
jameshart··on .gitignore Everything by Default
If your issue is that if you already have some changes to some files staged, and you have made subsequent changes to your working directory, if you git add . then it will not be possible to revert the cache to the state it was in before… well, yes, but there’s an easy way to checkpoint the state of the staged cache: git commit.

But to be clear: I’m the kind of person who runs git add .; that means I’m not maintaining a secret hidden third set of changes in the cache. I’m making changes in my working directory, and when those changes are in the shape I want them to be, I am staging all of those changes so I can commit all of those changes. If git add . overwrites a previous state of cached files, that’s fine - if I didn’t commit them I don’t want them.

Committing a codebase state that only exists in the cache and in your mind, on which you haven’t therefore run any build or test or linting tools because that precise set of changes has never existed in isolation in your codebase strikes me as requiring far too much mental powerlifting.

jameshart··on Actively exploited sandbox RCE in all Chromium versions
Ah the classic programmer mistake when encountering the legal system of thinking ‘but surely this act, in and of itself, could not be considered illegal?’ When what the law criminalizes are not mere acts but actions carried out with intent and knowledge in a context.

‘Officer, you can’t arrest me for carrying a gold bar!’

‘I can if you are carrying it out of the Federal Reserve vault without permission’

Similarly:

‘Surely selling someone the knowledge that a piece of software can be hacked isn’t illegal?’

‘It is if you know or suspect that that person’s intention is to use that knowledge to commit crimes’

This would make you a ‘coconspirator’ or possibly an ‘accessory before the fact’ in any crimes they committed.

jameshart··on Actively exploited sandbox RCE in all Chromium versions
The black market also prices in a risk premium for ‘this is illegal and you could go to jail for selling me this’.

Google is only paying for the vulnerability; the exploit market is also paying for your mortal soul.

jameshart··on .gitignore Everything by Default
git add . is nondestructive and reversible. I tend to do git add . and then run git status.

I’m far more interested in confirming my mental model of what I am about to commit if I git commit right now than my mental model of what will be added if I git add right now.

If I didn’t already have that muscle memory I might try to switch to git add -v . which would tell me what files it added. I’d argue that a sane git would just output git status after a git add operation.

But also a sane git would have a more logical command than git rm —cached to unstage a file. Like git unstage perhaps.

jameshart··on “Next-token predictor” is the wrong mental model for LLMs
Yep, we’re all just putting one foot in front of the other, hoping we’re doing the right thing to bring about the outcomes we want, trying our best.

But that’s the point: so is an LLM. Putting one token in front of another, hoping it’s doing the right thing to bring about the rewards it’s trained to… trying its best.

So yeah, not ‘next token predictors’. ‘Next token tryers’ maybe.

← PreviousPage 5 of 34Next →