frabcus

17 ms·

271.
by frabcus 3y ago
This was a huge thing in the 1990s - yes there is a singular definition, by the Open Source Initiative https://opensource.org/ That's a good thing, because otherwise corporations constantly try to stretch the definitio
272.
by frabcus 3y ago
Looking at https://www.mozilla.org/en-GB/firefox/developer/ I'm a bit confused... Which features are just new features which will soon be in the main Firefox DevTools, and which are specific to installin
273.
by frabcus 3y ago
Neuralink for cephalopods.
274.
Mirages: On Anthropomorphism in Dialogue Systems(arxiv.org)
41 points by frabcus 3y ago|22 comments
275.
What is high-quality about the data that trained generative AI?(flourish.org)
1 points by frabcus 3y ago|0 comments
276.
by frabcus 3y ago
I think it is more that language is a convenient way to capture high quality human intellectual capabilities to train an AI easily. LLMs have highly curated training text to keep that quality high. In reality we think not just in words but
277.
by frabcus 3y ago
Is there a MySQL plugin for vectors?
278.
by frabcus 3y ago
The vectors are usually (if you use OpenAI API anyway) unit in length, and so you can imagine them on the surface of a hypersphere. You measure the cosine distance between documents, or between search queries and documents. (Cosine is fast,
279.
by frabcus 3y ago
One example - they mine people's labour of searching beyond the first page of results - so when a small % of people really dig into results, they can infer which the good sites are deeper in (e.g. by when you settle on one). Bing doesn
280.
by frabcus 3y ago
Presume they'll bring the add-on into their standard monitoring / security checking! "This add-on is not actively monitored for security by Mozilla. Make sure you trust it before installing." https://addons.mo
281.
by frabcus 4y ago
LLMs are better at context - as far as I can tell Google Translate only does one sentence at a time. With GPT-3 it infers things like gender and formality from earlier context. Not clear yet how good GPT-4 is as OpenAI won’t say what traini
282.
by frabcus 4y ago
ELI5, how does this work? It claims to both identify humans and be zero knowledge. What’s to stop me registering a bunch of times then letting my bot use my identities? The answer is implied to be my iris scan. But then it isn’t zero knowle
283.
by frabcus 4y ago
It’s a bit more like a capitalist shoggoth mask on a non profit. Microsoft’s investment return is capped at 100x then they have no more claim. (Curious if they even have search engine exclusivity then!) So for now OpenAI acts capitalist, th
284.
by frabcus 4y ago
You get it to write them. Maybe in cucumber so you can check them / edit them by reading the English. Maybe you use a competitors model to write the tests as then less likely to make same error in code and tests, or write them twice an
285.
by frabcus 4y ago
I was told by OpenAI not to use the edit models - that the edit API is (effectively) deprecated. We were hitting a very low rate limit, so switched to using a completion model for the same thing (it's as good).
286.
by frabcus 4y ago
So presumably the model could output tokens that represent images as well? For multi-modal training data, e.g. HTML pages or PDFs, does the training data interleave the image tokens amongst the text tokens in the same document? Slightly l
287.
by frabcus 4y ago
You can get it to ask more questions…
288.
by frabcus 4y ago
According to the book Sapiens it depended on resources, but women often breast fed for 4 or 5 years partly as a way to spread out and reduce number of children. The need for lots of children was a consequence of farming and industrialisatio
289.
by frabcus 4y ago
In my experience both same time zone, and meeting up regularly (eg once a month) are important for making remote working as effective as possible. This can lead to a same country policy. If I was starting a new remote organisation now I’d a
290.
by frabcus 4y ago
It’s that your subconscious mind won’t feel safe, so you won’t unpack things to yourself that you would otherwise.
291.
by frabcus 4y ago
It more let’s the bot can query a database (using a DSL you teach it) when it wants to. So for example if it couldn’t remember an anecdote from the clients past, I guess it could search the history? eg ask to search for “Sarah" and the
292.
by frabcus 4y ago
API revenue so far (somebody has to be upstream provider to all the UIs to make AI marketing spam), gives them an advantage over AWS, although I’m sceptical they’ll keep it for long.
293.
by frabcus 4y ago
Out of interest are there any copyleft style neural network licenses - eg that require fine tuned model weights are published? (And Affero GPL style in terms of servers and distribution meaning these days)
294.
by frabcus 4y ago
My first guess is the human feedback used for training the Reinforcement Learning from Human Feedback (RLHF) network?
295.
by frabcus 4y ago
I think it was that click throughs (and someone say buying something) reduce ad impressions, as they’re no longer scrolling Twitter. As a result Twitter ad targeting has ended up only good for brand advertising, when at least in theory comp
296.
by frabcus 4y ago
There’s an interesting bit in the recording of Musk on Spaces the other day where he says it took him weeks to understand that the ad targeting has ended up targeting fewer click throughs, as that leads to more impressions. They’ve got into
297.
by frabcus 4y ago
In the recording of Elon Musk on Spaces the other day, there is a bunch of conversation about views being computed off prem in Google Cloud unlike, say, the likes count which is computed in the main Twitter in prem database. Wasn’t quite cl
298.
by frabcus 4y ago
That's quite funny! Behaviour when the prompt has contradictions. Since GPT-3 is a statistical model, what's the chance it calls out the question as wrong? Is there a reasonably high probility sequence which just says that? At som
299.
Language Models Can Teach Themselves to Program Better(arxiv.org)
7 points by frabcus 4y ago|0 comments
300.
by frabcus 4y ago
Because it is a solid line (the policy rule) drawn across a fractal boundary (actual porn), and given lots of attempts you can find somewhere inside the line but across the boundary. Stopping more than so many attempts makes this much harde

More ›