padolsey

18 ms·

271.
by padolsey 3y ago
I created https://a.foo a while ago which just summarizes URLs for me. Like, if I don't have time to read an article or something, I just pop 'a.foo' in front of the URI and it gives me a quick rundown. EDIT: semi
272.
by padolsey 3y ago
Quite interesting. So epileptic brains are re-using a mechanism from the sleep cycle to defend against daytime seizures [potentially].... But at a cost of cognitive ability. I like how the brain can adaptively rewire stuff to defend itself.
273.
Multifaceted: The linguistic echo chambers of LLMs(blog.j11y.io)
2 points by padolsey 3y ago|0 comments
274.
by padolsey 3y ago
This feels like it could be real remorse, and a true lapse of judgement based on good intentions. So, in the end: a story of Ilya, a truly principled but possibly naive scientist, and a board fixated on its charter. But in their haste, noth
275.
by padolsey 3y ago
So when the author states that "Microsoft just acquired OpenAI for $0" they mean, effectively, only a fixed-time snapshot of code that is likely old news in about 18 months by the time other models have caught up. Microsoft still
276.
by padolsey 3y ago
I'm a bit confused. Does MSFT have a perpetual license to the original OpenAI LLC's IP *or* the capped company OpenAI "Global" LLC that was specifically created for MSFT's stake? Because, if the latter, it seems lik
277.
by padolsey 3y ago
There are engineers who care about the kinds of values that OpenAI was founded on, which have just been – arguably – reaffirmed and revalidated by this latest drama. OpenAI's commercialization was only ever a means to have sufficient c
278.
by padolsey 3y ago
What makes you so sure? It feels like a GPT-5/6, with enough inner feedback loops and system-calling capability will be scarily capable.
279.
by padolsey 3y ago
If you're just using their completions/chat API, you're gonna be ok. As an ultimate fallback you can spin up H100s in the cloud and run VLLM atop a high param open model like Llama 70B. Such models will catch up and their par
280.
by padolsey 3y ago
Yeh I really wish they'd better articulate the "AI safety" directive in a way that is broader than deepfakes and nuclear/chemical winter. It feels like an easy sell to regulators. Meanwhile most of us are creating charmi
281.
by padolsey 3y ago
The biggest grudge I have against 'stroke' is that it's such an amorphous diagnosis. A heart attack, a similarly ischemic (or hemorhagic) event, often produces, say, one of a couple dozen lasting distinct symptoms. Some can b
282.
Using History Insertion, Policy Drift and Allusions to Jailbreak DALL·E 3(padolsey.medium.com)
3 points by padolsey 3y ago|0 comments
283.
by padolsey 3y ago
>I hate when someone says, "Oh, Schwarzenegger is the perfect example of a self-made man" because I'm not. I'm a creation of my parents. I'm a creation of my coaches, my teachers. I have been helped by my traini
284.
by padolsey 3y ago
I think this is kinda cool. It pushes forward our understanding of how to work with LLMs. However, it doesn't appear to be something that can be relied upon idempotently. It seems liable to every flaw people have identified in LLMs so
285.
by padolsey 3y ago
The real news here is the uncapped indemnification and protection from any copyright lawsuits. This is a good move from Getty.
286.
by padolsey 3y ago
Not to minimise OP’s work, but if you’re looking for markdown in a single chat thread, you usually can just ask chatgpt to give you markdown in a code block.
287.
by padolsey 3y ago
Cheeky and it probably won’t work. I initially thought it’d be a crowdpooling thing where people could sign up to a pool of distributed/tunnelled copilot users. That would be cool… Lots of copilot subscriptions are sitting idle at any
288.
by padolsey 3y ago
I had the exact same feeling several years ago so I made a graph-based app that does that! ablf.io/?q & ablf.io/goodreads (the latter lets you upload your existing goodreads lib if u want)
289.
by padolsey 3y ago
Summary: Investors don't know where to put their money and they're scared
290.
by padolsey 3y ago
Yeh it's bizarre. I guess, inevitably, he is viewing everything through a crypto-centric mental model of the world. Round-peg-square-hole thinking. Community Notes are not truly permissionless, trustless, or decentralized. The algorith
291.
by padolsey 3y ago
Oh yeh definitely. I meant more specifically: might it be possible to capture the electrical signals (much like this current system) from the parts of the motor cortex that create the series of muscle movements that form a 'sign',
292.
by padolsey 3y ago
> Rather than train the AI to recognize whole words, the researchers created a system that decodes words from smaller components called phonemes. These are the sub-units of speech that form spoken words in the same way that letters form
293.
by padolsey 3y ago
Oh hm true - I guess I meant tobacco. Whoops!
294.
by padolsey 3y ago
It’s true. But also perhaps more nuanced. Many people see disability as a purely negative experience, something to avoid at all costs, synonymous with illness. But it is a slightly different axis: Both medical and societal in nature. Thus w
295.
by padolsey 3y ago
Indeed. And we are all born disabled (in that we are unable to sustain our own independence and wellbeing without assistance), and will likely all be disabled at the door of death. For such a universal experience, you’d think healthcare wa
296.
by padolsey 3y ago
This is quite darkly hilarious tbh. And consistent with my experience of Blackpool. I once went there on couple-week-long driving course when I was a teenager, hoping but eventually failing to get my license. I was put up in someone's
297.
by padolsey 3y ago
I reckon it's likely that biases emerge more from the virtues that have been encoded in the interfaces than from the models themselves. So any identified political slant isn't very interesting IMHO. ChatGPT and others are understa
298.
by padolsey 3y ago
Ostenibly yeh JSON should be able to encapsulate mose of that semantic stuff but having replaced an XML schema in the system prompt with gpt's function-calling API I've been very umimpressed. It feels much less capable. I would ha
299.
by padolsey 3y ago
XML Elements themselves: their naming, their attributes, comments, indentation. There's more opportunity at every level of the hierarchy to demarkate and establish meaning. Having closing-tags as well, I've found, is a massive boo
300.
by padolsey 3y ago
IANA{LLM}, but if you're only sampling from a "correct" grammar, you are potentially (very potentially) forgoing what might otherwise have been a more desirable and more semantically useful token. Most of the models have been

More ›