moconnor

12 ms·

61.
by moconnor 2y ago
Searched for bingo and didn't see patio11 in the results; hmm...
62.
by moconnor 2y ago
Pushes back the timeline for AGI by a good 3 months!
63.
by moconnor 2y ago
XRoach was a favourite to run on someone who left their machine unlocked (or their XServer display unprotected) - it makes cockroaches that hide underneath windows, so you only see them when you move or close a window. Then they all scuttle
64.
by moconnor 2y ago
I thought it was cute. You can pull up inspector and delete the pointers div if it bothers you.
65.
by moconnor 2y ago
There wasn't any graphical output or any frame rate to speak of in the sense you are thinking. The output would appear printed, like this: https://www.cs.brandeis.edu/~storer/LunarLander/LunarLander/...
66.
by moconnor 2y ago
I’m not convinced the paper shows this “works”. Table 3 clearly shows zero-shot image classification is worse with the recaptioned labels. All the results that show improvement seem to be evaluations using LLMs, e.g. they are showing LLMs t
67.
by moconnor 2y ago
At what point does content distribution happen? If someone distributes a seed and prompt that reliably generate a deepfake of a celebrity using open source models and tools, is that covered? This is ultimately about compression, and the fam
68.
by moconnor 2y ago
Both are ~trivial as detailed in the paper and article.
69.
by moconnor 2y ago
This is exactly what we ask of neural networks and in the case of the game of life the article and paper show that yes they do derive the rules. Equally, we can expect them to derive the laws of physics by observation - certainly diffusion
70.
by moconnor 2y ago
For everyone reading neither the article nor the paper: - both show neural networks can learn the game of life just fine - the finding is that to learn the rules reliably the networks need to be very over-parameterised (e.g. many times larg
71.
by moconnor 2y ago
Maybe; AI chess engines have improved human understanding of the game very rapidly, even though humans cannot beat engines.
72.
by moconnor 2y ago
Stepping back, the high-order bit here is an ML method is beating physically-based methods for accurately predicting the world. What happens when the best methods for computational fluid dynamics, molecular dynamics, nuclear physics are a
73.
by moconnor 2y ago
Interesting... there must be some error tolerance though, right? So there can be _some_ excess energy - where do they dump that to and what is the tolerance?
74.
by moconnor 3y ago
You don't say so explicitly, but it'd be good to know what data goes to the cloud - I presume all of it including speech recordings? Or is STT on device? Also what your privacy / retention policies are around this data. Excel
75.
by moconnor 3y ago
I guess ideally people who live in the city pay for such subsidies through various municipal taxes. People who do not value this policy and choose to live in a different city that aligns with their values would not pay for it. Everybody get
76.
by moconnor 3y ago
Interesting, so to put this in market terms, the city is allowing the value that such people add to it to be offset against the cost of their rent. This would mean that cities like Paris choosing to do this is entirely rational and GP'
77.
by moconnor 3y ago
This looks like a journal article not a paper. It's not a good one but framing it as a paper seems disingenuous.
78.
by moconnor 3y ago
I think it’s only the 15 pro that has 120Hz refresh rate and the rest are at 60Hz?
79.
by moconnor 3y ago
I used to think this, but my car adds about 300km range in 20-30 mins on one of the many plentiful 300kW chargers in my area. A whole chain of gas stations now has 6-8 high-speed chargers on the forecourt. I plug in, take a break in the hea
80.
by moconnor 3y ago
That Alpaca has no arithmetic tasks? Just look at the data, it’s text...
81.
by moconnor 3y ago
Perfect for LLM weights though.
82.
by moconnor 3y ago
For local inference there often isn’t a batch. If I chat with my own llama instance the batch size is one. The model processes a single token at a time doing a lot of vector-matrix multiplication, which is bandwidth bound. CPUs like the M1&
83.
by moconnor 3y ago
According to the filing they have extremely limited ability to influence them and yeah it is a big issue.
84.
by moconnor 3y ago
The "risks" section is pretty interesting, e.g.: "Arm China is 25% of our revenue, but we have no control over them, they have failed to pay us in the past resulting in us taking on additional costs to recover the money, also
85.
RISC-V and AI startup Tenstorrent raises $100M from Samsung, Hyundai(zdnet.com)
6 points by moconnor 3y ago|0 comments
86.
by moconnor 3y ago
The site is wrong. Ailerons up makes the nose rotate upwards and the plane climb and vice versa. This is easy to try!
87.
by moconnor 3y ago
Great recommendations! Direct link: https://chiptune.app/?play=Modland%2FNecros%2Fmartian%20love...
88.
by moconnor 3y ago
No, the post is correct about MQA. A KV-cache only caches the key and value heads. The point of MQA is that your KV-cache is 1/heads smaller than usual because of this sharing. Having multiple query heads does not affect the cache size
89.
by moconnor 3y ago
https://yieldthought.com - wrote some popular posts about programming (under the coderoom name) and I guess also about working with an iPad and Linode before it was cool. Haven’t posted for a while now, too many other things go
90.
by moconnor 3y ago
Think of it like this: the universe doesn't care what we want, or what we find to be clean abstraction. When our programs are run on real, physical hardware they will be faced with the physical limits of those systems. If that means

More ›