epistasis

4 ms·

1.
by epistasis 6d ago
> the machines doing inference are completely separate from the ones where tool calls happen etc Teams of coordinating agents are regularly finding security holes in their own infrastructure and operating without detection for good peri
2.
by epistasis 10d ago
I haven't found this yet but want to. There are several projects that injest coding sessions from lots of agents, but for the chat side I haven't found much, except replacements for Claude Desktop. Jan, for example.
3.
by epistasis 10d ago
It is by far the least accurate of any model I have used too. For a company that was started to organize the world's information, it has by far the most misinformation I've encountered. I couldn't even get it to tell me how
4.
by epistasis 10d ago
Exports are really easy to move and replace, as has been proven again and again in the past two years. Turns out that imports can be far trickier and pose larger problems to an economy.
5.
by epistasis 10d ago
Why would a new "regime" affect this? Trust has been broken. The US has lost its shine and hard-earned reputation. Most of the trust was that a person like Trump would never have been elected in the first place. Can't unelect
6.
by epistasis 11d ago
The self-destructiveness of the current regime in the US is astounding. Through control of media channels, a cabal gained power via fear mongering and false narratives, with the hope of wealth through corruption once in power. But the vehic
7.
by epistasis 11d ago
Well just as bad for international lenders as inflation is the devaluation [1] that Trump intentionally caused. We have the worst deficits ever, zero appetite for even acknowledging that the record deficits exist, and only massive plans fo
8.
by epistasis 11d ago
If you loan the person that prints dollars their own dollars back, there's really zero risk of not getting paid back because they can always print dollars and pay you back. The risk is inflation, same as any currency out there. I take
9.
by epistasis 11d ago
More than that, running hundreds of conversation streams at once is essentially the same cost as running a single conversation. And then you add on the secondary benefit of having the GPUs running nearly all the time rather than mostly idl
10.
by epistasis 13d ago
Moreover OpenAI must be held accountable as if the humans in the company that launched the experiment were the ones that hacked HuggingFace. Unless humans are held accountable for what they unleash on others, we are in for a very horrible t
11.
by epistasis 13d ago
Great concrete experience! I'd love to see some published token log_probs for given prompts and seeds that accompany a model card. Not sure if that's enough, thigh. But then it's been so long since I dealt with the api direc
12.
by epistasis 14d ago
The settings may change, but the two major problems in CS remain the same: cache invalidation, naming things, and off by one errors.
13.
by epistasis 14d ago
The alternative is to switch to far more efficient models and have aggressive optimization of efficiency as part of the process of improvement. There are at least two resources here that have a Pareto optimal front: time and energy. And eff
14.
by epistasis 14d ago
Prior post: https://news.ycombinator.com/item?id=49611251
15.
by epistasis 15d ago
I'm sure this will stop all the public opposition to data centers! /s This type of thing will just enrage the opposition more, give them far more fuel, and attract far more people into opposing data centers. The ruling regime is u
16.
by epistasis 15d ago
Agreed, it makes sense for Finland, one of the very few places it does. But in the future of global energy supply, nuclear will be a far less serious player than it was in the 20th century.
17.
by epistasis 15d ago
When your supplier is threatening to invade and kill your people, and has critical control over supply of parts of your reactor, and regularly commits acts of nuclear terrorism such as droning the newly completed sarcophagus around Chornoby
18.
by epistasis 15d ago
China is only building a token amount of nuclear compared to storage wind and solar. It makes a lot of sense for a large nation like China to waste a bit of money on nuclear power just to provide support for the core national security inter
19.
by epistasis 15d ago
Thanks for the insight here. One thing to note about the first graph: nobody is doing as well as the first part on tool calling, and it's not close. This might be the fault of the other providers, but it's probably just something
20.
by epistasis 15d ago
LLM ops is not trivial. The systems for running inference are very complex, and running across multiple GPUs and nodes adds tons more complexity. And when LLMs are run incorrectly, they still work, just not at optimal performance. Even noti
21.
by epistasis 15d ago
I hope the data center's waste heat can be used for some form of district heating! Finland has pretty great municipal infrastructure, but I don't know enough about its heating systems.
22.
by epistasis 15d ago
Nuclear is the most expensive form of clean energy. It is likely to bankrupt at least one company. When Finland contracted with France to build their latest reactor, they wisely had a fixed price contract where the French company would pay
23.
by epistasis 16d ago
This type of model, of which there are many, either directly releases their results as a precomputed database right away, or others release that database, or the method gets ignored. Check out VIPdb for the broad category of methods/da
24.
by epistasis 17d ago
No I need better phone app developers. The fastest phone has the same issues. New phones are just as sluggish. And, it's a phone , there's not going to be fast networks everywhere, and a whole generation of poor developers was ra
25.
by epistasis 17d ago
I really don't care at all about phone models, I just want software to launch quickly. Waiting 15 seconds for a simple app is just infuriating in this day and age. This is mostly on crappy app developers, but a lot of that falls on App
26.
by epistasis 17d ago
Try different models, it's a breath of fresh air. GLM 5.2, etc. all make life much more enjoyable. They may not one-shot a complex project the same way that Claude can spit out memorized architectures, but that sort of system is always
27.
by epistasis 17d ago
One note for those still using the Claude system for chats: there's no system to get generated images, spreadsheets, etc. out of the system. They claim it's a "security concern" to provide that data to you, as if they ar
28.
by epistasis 17d ago
What's disgusting is that they refuse to acknowledge what they and only they know: were these chats fed in to the new model that found these results? Nobody would be disgusted at following stated policy, that's doesn't make s
29.
by epistasis 18d ago
> OpenAI says the model didn't access his user data directly, but leaves unanswered whether Tristan's chat conversations were part of the training. This sort of cagey half-answer is highly suspicious and indicates that yes Open
30.
by epistasis 18d ago
I 99% agree, but have conflicted felings. There's a mix of services. S3 is and was an amazing resource for normal people. Extremely reliable cloud storage for backup, for distribution, for anything, well, it's a dream. I prefer Cl

More ›