bjornsing

7 ms·

1.
by bjornsing 15d ago
Location: Malmö / Stockholm, Sweden Remote: Yes Willing to relocate: No Résumé/CV: https://www.linkedin.com/in/bjornsmedman/ Email: bjornsmedman [at] gmail.com Engineering-physicist with 20
2.
by bjornsing 7mo ago
The concept is called static analysis.
3.
by bjornsing 11mo ago
Will Anthropic/OpenAI really hire anyone who can fine-tune an LLM?
4.
by bjornsing 11mo ago
I ran a small ISP around the same time that used this behavioral pattern to bring down the customer acquisition cost to near zero. Essentially we sold ADSL connections with Wi-Fi and a second SSID where anybody could connect and sign up for
5.
by bjornsing 11mo ago
To some extent I think that’s just human nature, or even animal nature. The optimal explore / exploit tradeoff changes as we age. When we’re children it’s beneficial to explore. As adults it’s often more beneficial to exploit. But you
6.
by bjornsing 11mo ago
Yeah that’s sort of how I understand the OP too: The CPU will execute speculatively on the assumption that the next element in the linked list is consecutive in memory, so it doesn’t have to wait for L1 cache. It needs to check the real val
7.
by bjornsing 11mo ago
Yes. But I don’t think the OP is suggesting this as an alternative to using an array. As I read / skimmed it the linked list is just a simplified example. You can use this trick in more complex situations too, eg if you’re searching a
8.
by bjornsing 11mo ago
But that works on a different level, right? At least as I understand it data speculation is about prefetching from memory into cache. This trick is about using the branch predictor as an ultra-fast ”L0” cache you could say. At least that’s
9.
by bjornsing 11mo ago
Yeah I think this is a general principle. Just look at the quality of US presidents over time, or generations of top physicists. I guess it’s just a numbers game: the number of genuinely interested people is relatively constant while the nu
10.
by bjornsing 1y ago
Yeah the whole methodology depends on forgetting about state and treating it as a long-running program. If you need to look at the state then you connect a debugger, etc.
11.
by bjornsing 1y ago
Sure. But that’s just compression, right? I guess you could argue that some information is stored outside the genome, in the structure of proteins etc. But the counter argument is that that information is quickly lost in cell divisions. Onl
12.
by bjornsing 1y ago
Swedish banks (even the Riksbank linked above) regularly refuse to turn cash into digital money unless you can ”prove” where you got it from. It’s not sufficient to say (with immense credibility) that you worked hard all your life and saved
13.
by bjornsing 1y ago
You’d have to explain where that innate knowledge is stored though. The entire human genome is less than a GB if I remember correctly. Some of that being allocated to ”priors” for neural circuit development seems reasonable, but it can’t be
14.
by bjornsing 1y ago
The challenges around idempotency remain to some extent, yes. But you have that problem even in non-workflow code, so the usual patterns will just work with no extra mental effort from the developer.
15.
by bjornsing 1y ago
Sure you get more control with explicit state management. But it’s also more work, and more difficult work. You can do a lot of writes to NVMe for one developer salary.
16.
by bjornsing 1y ago
"Exactly-Once Event Processing" is possible if (all!) the processing results go into a transactional database along with the stream position marker in a single transaction. That’s probably the mechanism they are relying on.
17.
by bjornsing 1y ago
> Yes, in any durability framework there's still the possibility that a process crashes mid-step, in which case you have no choice but to restart the step. Golem [1] is an interesting counterexample to this. They run your code in a
18.
by bjornsing 1y ago
But couldn’t an LLM search for documents in that enterprise knowledge base just like humans do, using the same kind of queries and the same underlying search infrastructure?
19.
by bjornsing 1y ago
Exactly.
20.
by bjornsing 1y ago
I doubt it was due to network effects. I’m probably one of those top quartile potential H1-B holders that never applied. My main reason for not applying is that the random nature of the lottery reduces the RoI on finding a US job so much th
21.
by bjornsing 1y ago
I guess it needs to get across the blood-brain barrier. But that shouldn’t take 10+ hours of surgery, I don’t think.
22.
by bjornsing 1y ago
But it doesn’t take 10+ hours to surgically get a virus across the blood-brain barrier, right?
23.
by bjornsing 1y ago
Why does it have to be delivered through brain surgery?
24.
by bjornsing 1y ago
I’d say an LLM is a form of Markov chain, where the state is defined to consist of the entire context window, and each state transition adds one token to that context window.
25.
by bjornsing 1y ago
> So for data centers, training is just as important as inference. Sure, and I’m not saying buying Nvidia is a bad bet. It’s the most flexible and mature hardware out there, and the huge installed base also means you know future innovati
26.
by bjornsing 1y ago
> CUDA has been a huge moat The CUDA moat is extremely exaggerated for deep learning, especially for inference. It’s simply not hard to do matrix multiplication and a few activation functions here and there.
27.
by bjornsing 1y ago
> While that doesn’t mean their cosine distance is large There’s a lot of devil in this detail.
28.
by bjornsing 1y ago
I agree the OPs argument is a bad one. But I’m still optimistic about the representational capacity of those 20k dimensions.
29.
by bjornsing 1y ago
Good read. I’ve thought about this distinction myself, but this was definitely more clearly articulated. > This view is kind of like saying that engineers are just people who weren’t smart enough to do physics, or physicists aren’t smart
30.
by bjornsing 1y ago
So how do you best store and iterate on the spec? One way I guess would be to work on a branch an modify Claude.md to reflect what the branch is for. Is that a good approach? Are there others?

More ›