HNHacker News
TopNewBestAskShowJobs

blueboo

793 karma · joined May 9, 2018

submissionscomments
blueboo··on My kid sounds like ChatGPT, and soon yours might, too
Conspicuous absence of evidence or even specifics of what the author thinks “sounding like ChatGPT”. I personally wouldn’t characterise it as “robotic” or akin to a trope-y robotic character (Jane from the Good Place).

But credulously taking claims at face value, structure and clarity is a good thing for a teenager’s expository writing.

blueboo··on Japan to crack down on Apple and Google app store monopolies
Agreed, mostly, as Id like to point out that the iOS software is abandoned/shut down at approximately the same cadence as new consoles are launched. Rare’s the still-available app that was last updated pre-iOS 10…let alone iOS 5 or pre-retina iOS.
blueboo··on Tesla Drivers Have the Highest Crash Rate of Any Brand: Study
Punishing “poor slobs” is the cornerstone of deterrence for most crimes, why doesn’t it apply here as well?
blueboo··on Word2Vec received 'strong reject' four times at ICLR2013
This is interesting. I went off and searched for paragraph vector code and indeed find doc2vec stuff, including tutorials referring to the paper such as https://radimrehurek.com/gensim/auto_examples/howtos/run_doc.... It’s not obvious that the results aren’t reproducible (and I realise code is not the same as published results), but I wonder if you could steer us more specifically.
blueboo··on OpenAI suspends ByteDance's account after it used GPT to train its own AI model
“Simply”? It would be quite a project to do that at sufficient scale undetected, no?
blueboo··on Pipe Dreams: The life and times of Yahoo Pipes
Laggily zooms a bit…then breaks on iOS becoming a big scrollable white page. I bet when it works the animation is kind of neat.

So much work to bounce your readers, pity!

blueboo··on Sorry, but a new prompt for GPT-4 is not a paper
I’m not sure the scientific method itself can withstand this sort of scrutiny. After all, it’s just making guesses about what will happen and then seeing what happens!
blueboo··on Are Open-Source Large Language Models Catching Up?
Which is why there are 100+ LLMs in China ... the so-called 百模大战, battle of the 100 models.
blueboo··on Apple and Google avoid naming ChatGPT as their 'app of the year'
I think a strong argument can be made in the other way. I went running the other day and had a long conversation with ChatGPT. That sort of experience is absolutely pushing boundaries...but I can see that they aren't the boundaries that Apple wants to celebrate at this moment. Apple has set a pattern of rewarding beautiful, self-contained, cohesive, through-designed, snackable experiences. A technology that challenges the notion of a smartphone itself is basically out-of-bounds.
blueboo··on Meta disbanded its Responsible AI team
An “innovation” team is often useful…usually it’s called research or labs or skunkworks or incubator. It’s still terrifically difficult for a large company to disrupt itself — and the analogy may hold for “responsibility”. But there is a coherent theory here.

In this case, there are “responsibility”-scoped technologies that can be built and applied across products: measuring distributional bias, debiasing, differential privacy, societal harms, red-teaming processes, among many others. These things can be tricky to spin up and centralising them can be viable (at least in theory).

blueboo··on New York may ban noncompete employment agreements and Wall Street is not happy
If it doesn’t make sense in California’s Silicon Valley, how can it be justified in NYC
blueboo··on We Automated Bullshit
Not sure if your meme-quippery works. Non-drunk driving kills people too, and also helps people get to work on time.
blueboo··on Copilot for Docs
What is copilot for docs? This page, rather infuriatingly, doesn’t explain! Is it part of copilot? A product? An experiment…?

Copilot for docs is ___

blueboo··on Leaders Are Tool Builders: Why I Wrote My Own JavaScript UI Framework
I think the title is another sensible variation of tech leads as workflow-unblockers, corner-uncutters and process-robustifiers. For all of these functions, however, you still need good judgment.

I do wonder what exactly this particular choice, a micro framework, unblocks? I’m reading some broad concerns about the provenance of React but it’s not clear that’s grounded in a need of the team.

There are some challenging dynamics on the horizon for this team…I’d be keen to hear how this goes, six or twelve months on.

blueboo··on Grok is an AI modeled after the Hitchhiker’s Guide to the Galaxy
There are all sorts. You do have to be specific about what you mean by “bias” though. https://arxiv.org/abs/2110.08193 and https://arxiv.org/abs/1804.09301 and https://arxiv.org/abs/2206.04615 …
blueboo··on How to Not Get Screwed over as a Software Engineer [video]
I worry this whole comment chain is celebrating a coping mechanism of troglodytic engineering, that the ideal is to shape artisanal code-crystals undisturbed in a cave. We all wish there was a simple meritocracy of crystal-fashioning, with the master diamondsmiths sat at the top. If only, enough said.

The political savviness so scorned by the in-the-trenches coder might actually be communication, capacity for broader context, strategic foresight, emotional intelligence to debug and optimise teams, among the many other skills required to execute a product/project beyond programming. And good leads will pluck coders showing these skills out of the engineering corps and set them on a TPM track.

Perhaps it is only rarely the case. But it seems fair to observe that it is gentler on the ego to cast these rare and hard-earned skills as “playing the game”, PM mumbled-jumbo, etc.

At the same time, operators in bad faith tend to deploy just these skills, and a bad experience can incline one to wash their hands of it all and crave the rough diamonds of the cave…

blueboo··on Is supervised learning dead for computer vision?
We were always in this regime. In general, it has been far more effective to take a proven model and fine-tune it. It was true five years ago taking a convnet out of a model zoo to fine tune and it’s true now.

If you can achieve the moonshot of gathering, generating, annotating enough data with the right distribution to train from scratch, and you use a SOTA bag of tricks to regularise, you might do better.

Bear in mind fine tuning is literally just more pre-training. Starting with a trained model is like starting with an incredibly well-initialised network.

blueboo··on Microsoft: Require user consent before sending any telemetry
Yes, but that’s our childish instinct to be affronted at being held to account for what we know we’re responsible for.
blueboo··on Microsoft: Require user consent before sending any telemetry
The current system highly incentivises sufficiently large corporations to embrace the Nike principles: Break the rules, fight the law

The worst case scenario, if you lose a game stacked in your favour several times in a row, you pay a pittance, or performatively correct a now-obsolete injustice.

VScode telemetry will remain opt out because it yields very valuable information. Microsoft is not a democracy, and the outcry here is less than a rounding error, a footnote in some internal director’s morning agenda.

blueboo··on Ask HN: Why has no one replaced Ticketmaster?
Bigger fish to fry

Ticketmaster is big but it’s not Amazon or banking.

blueboo··on AI Reduces the World to Stereotypes
It’s both — the average depiction tends to be stereotype-based
blueboo··on How the creator of Alone in the Dark came back to games
Many? In 92? I don’t think so. AitD was the first in my (admittedly non-comprehensive) experience. 92 was full of sprite-based pixel art adventures, to be sure.

Relentless/Twinsens was 94. RE and FF7 were years after that

blueboo··on The Geometry of Truth: Do LLM's Know True and False
True, particularly when you consider what fMRIs tell us about a brain’s state of knowledge
blueboo··on The Killer Use Case for LLMs Is Summarization
My favourite extant example of this GPT4’s 18-way rule-based reward model classifier, appendix A https://cdn.openai.com/papers/gpt-4-system-card.pdf
blueboo··on Tiny Language Models Come of Age
Believing this is existential for a thousand AI seed-stage startups. I hope it’s true for their sakes.

Hear the other timeline from Darius Amodei. https://a16z.com/improving-ai/ We may be entering a regime of multi-billion training costs on custom hardware. Open-source AI will be like open-source CPUs. Cool but the real thing even at the smallest scale comes out of monopsonized (is that a word?) inaccessible infrastructure

blueboo··on What Is Appropriation in Art?
“Appropriation” has been a technical term in art criticism distinct from “cultural appropriation”. “Cultural appropriation” which is what “appropriation” typically refers to in today’s mainstream discourse.

The art term is neutral. The cultural term is negative. Modern and contemporary art is plausibly rife with cultural appropriation, often coinciding with typical artistic “appropriation”, and this essay doesn’t engage with that issue, and its lack of acknowledgement of the imminent confusion is puzzling.

blueboo··on AMD may get across the CUDA moat
Research kernels mostly turn to ash upon publication anyway. The wheel turns and the next post-doc gives ROCm a try and we move on
blueboo··on Mistral 7B
Persimmon-8B from Adept did it first no https://www.adept.ai/blog/persimmon-8b
blueboo··on Gaussian splatting is pretty cool
There’s no helping the incurious, but there are indeed reams of explorations unto weights and their explanations.

One of my personal favourites is Zeiler and Fergus https://arxiv.org/abs/1311.2901

Tom Zahavy’s “Greying the black box” on deep RL comes to mind too https://arxiv.org/abs/1602.02658

OpenAI has their Microscope tool for investigating individual neurons https://openai.com/research/microscope; more recently OpenAI shared research and tooling leveraging a powerful language model to explain a smaller model’s neurons. https://openai.com/research/language-models-can-explain-neur...

For a true toy box exploration playground.tensorflow.org is worth a play

These may or may not satisfy your curiosities — your perfect tool may not exist…yet. I would argue that the sampling of extant work above are indicative that it and more are possible.

blueboo··on Wet-bulb temperature
“All rebels are closet aristocrats. That’s why I can convert them so easily”

Leto II, God Emperor of Dune, Frank Herbert

← PreviousPage 6 of 20Next →