andy_ppp

5 ms·

1.
by andy_ppp 8d ago
Except Taalas is much faster than GPUs, orders of magnitude so. They aren’t going to get 100x faster at inference any time soon!
2.
by andy_ppp 9d ago
Not sure, the labs will probably just cripple the security features of these models for a while I think and even potentially put back doors into systems for the security services…
3.
by andy_ppp 15d ago
I've never found feature flag systems complex and you can build one yourself in a few functions. The services I have used in the past were extremely simple to integrate so I don't understand the issue either way honestly.
4.
by andy_ppp 18d ago
The referendum was done specifically in two parts - 1. do you want us to try to find a deal with the EU 2. We will thrash out any details, get the best deal we can with the EU and give you another vote. Seems like Iceland didn't want t
5.
by andy_ppp 19d ago
Most people do not enjoy the output of the open models or know where their code is being sent so I’m out. Also I don’t trust Elon anyway, I’m just going to make a VS Code Extension that handles things how I want but I’d like to offer the sa
6.
by andy_ppp 19d ago
Why you would trust Elon Musk to not hack you and steal your model weights is beyond me, but maybe they intend to use that compute just for training, still I’m sure there’s plenty of research there to take too…
7.
by andy_ppp 21d ago
So did they train the AI to be superior to everyone and make bad decisions and never ask for clarification?
8.
by andy_ppp 22d ago
No, you were deliberately disingenuous (I understood that) but I decided to answer the best possible interpretation of what you meant. I'd seriously reconsider this argument style, just because it's common doesn't mean any of
9.
by andy_ppp 22d ago
Nonsense, you can fast or live in ketosis for years without any adverse consequences.
10.
by andy_ppp 22d ago
Yes, they could also sell me GPT Sol 5.6 or 5.7 on a chip and I’d probably buy it. It’s a really really useful model for me, I’m not sure how much better for coding I need it to be. For most things I find Sol good enough with a small amount
11.
by andy_ppp 22d ago
I’m willing to bet we can see longevity improvements from better blood sugar control. Not 100% certain but extremely likely, this would include potential heart attack and dementia reduction…
12.
by andy_ppp 23d ago
Yes, if it cannot be measured it doesn’t exist no matter how much it would make everyone feel more comfortable if we could intuitively understand these systems.
13.
by andy_ppp 23d ago
The original RoPE paper is a great example of the authors not actually understanding how their own discovery was functioning fully which is a fun example of what I’m talking about I guess… Yes you might progress the field, but will you real
14.
by andy_ppp 23d ago
I think we are talking about different things to be honest. Understanding LLMs is a fine thing to do but to say you understand what the network is actually doing, even in toy examples like MNIST is extremely tricky. Figuring out what a ~3 t
15.
by andy_ppp 23d ago
What provable conclusions have you intuited around how LLMs work? Give me some examples to prove your point? I'm absolutely happy to change my view with enough data.
16.
by andy_ppp 23d ago
I'm not sure it's possible to have intuition about systems that work in thousands of orthogonal dimensions. In fact I'm pretty sure most of the research is people trying fairly arbitrary things and testing them and then post
17.
by andy_ppp 24d ago
I bet you 100% it was built not by in house team but farmed out to an agency for £££££ millions. They will have been saying in their case studies some absolute nonsense about how they saved the government millions rather than burned loads o
18.
by andy_ppp 24d ago
How many coconuts do we need to support current aviation needs? I’m guessing it’s not likely this will work economically?
19.
by andy_ppp 25d ago
Interesting project feel free to share?
20.
by andy_ppp 25d ago
Yes same also cancelled my subscription, poor quality and slow.
21.
by andy_ppp 25d ago
I was thinking how tricky it would be to learn to program now, honestly a sense of laziness is one of the reasons I learned programming - so that I could automate away repetitive tasks. Our generation might be the last people who can actual
22.
by andy_ppp 26d ago
Can the US even afford a trade war, they are beginning to find borrowing more money increasingly difficult… currently the debt is around $130k for every man, woman and child in the US.
23.
by andy_ppp 27d ago
Until you get an event (or 104561 events) that corrupts the whole system and makes it really tricky to get back a working version.
24.
by andy_ppp 27d ago
Option 1, you’re filtering Claude with a local model using Anthropic’s API Option 2, you’re filtering what gets to ChatGPT with OpenAI’s API… I’m slightly confused :-/ does Anthropic off an Open AI compatible API?
25.
by andy_ppp 27d ago
Yes, you know why - it had nothing to do with copyright infringement - he was going to win the fight to get money out of politics.
26.
by andy_ppp 27d ago
From the README.md > Anything that uses the OpenAI API? I would have thought they meant the Anthropic API or maybe I'm misunderstanding?
27.
by andy_ppp 28d ago
I often think adding minority report style gestures to a laptop would be more useful than a touchscreen, maybe that’s just me…
28.
by andy_ppp 29d ago
1. If you get banned from Amazon largely this is very bad for business. They can find a reasonable excuse I’m sure! 2. Fraud is a stretch, these things are labelled even if it’s deliberately designed to be difficult to understand. The disco
29.
by andy_ppp 1mo ago
You will feel better and be in hospital at the end of your life for a lot less time on average so yes it’s a lot but potentially the chances of being a fit 80 year old could be worth it.
30.
by andy_ppp 1mo ago
Such good advice. Another probably unpopular use of AI - I was using Claude to figure out all the music I listened to at university and what would have been playing in the bars at the time. I was loving the nostalgia and I would not have re

More ›