infogulch

4 ms·

1.
by infogulch 6d ago
There's little credible threat that LLMs can actually upload their weights given that the machines doing inference are completely separate from the ones where tool calls happen etc. Weights are encrypted and locked on to the GPUs etc a
2.
by infogulch 7d ago
Thanks! The build system is nix, but the result is more of an appliance than a distro. (linuxManualConfig from tinyconfig + fragment, static musl Rust PID 1, mke2fs -d) Updates are rebuild-redeploy with the provider API; it can't updat
3.
by infogulch 8d ago
I made my own tunnel system with a $5/mo vps that runs kernel wireguard and accepts my nas' public key. Once connected it DNATs 80/443 traffic down the tunnel to the nas, where its routed to caddy. The vps runs a custom image
4.
by infogulch 9d ago
Yes it's clear that Sam and Dario seem to be aligned with a value set that prioritizes concentrating trans-national government-mandated centralized control of AI and crowning themselves high priests of this unholy abomination. "At
5.
by infogulch 9d ago
So they get down from 1.58 to 1.48 bits per weight by exploiting the fact that actual weights in practice are 0 51% of the time. Neat. If ternary llms work out and are baked into hardware as custom silicon I bet they'll be shockingly e
6.
by infogulch 13d ago
To kill someone with a gun a person has to pull the trigger with the clear intent to kill. The concern with AI, and the purpose of the OG paperclip maximizer thought experiment, is that the outcome could be extremely divergent from the inte
7.
by infogulch 13d ago
That's not a denial, that's confirmation that they are spying on you.
8.
by infogulch 18d ago
Can we imprint Claude with a little more Antoine de Saint-Exupéry: > Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away.
9.
by infogulch 18d ago
If you want to stick with a pixel maybe skip the Pixel 11 if you want to run GrapheneOS. There is some question whether P11 will actually support the MTE cpu extension which GOS requires for all devices it supports. Last thread about it: h
10.
by infogulch 22d ago
I made an MCP for myself that syncs all my Microsoft mail into a local sqlite db and exposes it as read-only to the agent. `query` tool can do search (FTS5 match) or browse (by date) on mailboxes/threads and returns lists of email with
11.
by infogulch 23d ago
Posted by @SpaceXAI 1h ago: https://x.com/SpaceXAI/status/2095597264043717014 > We are sorry for the issues you may have experienced with Grok following an outage at our Memphis compute center this morning. We’
12.
by infogulch 24d ago
This isn't a quality control issue, but a beancounter-chooses-cheaper-parts issue.
13.
by infogulch 25d ago
What is ERSC?
14.
by infogulch 1mo ago
Modern authorization systems are often graph-based. Check out ReBAC and ABAC authorization models and also implementations like apache/casbin or authzed/spicedb. These schemas often have surprisingly simple graph definitions; I be
15.
by infogulch 1mo ago
Sounds fixable.
16.
by infogulch 1mo ago
Last month a Show HN: ContextVault proposed an interesting long-term memory architecture. I discussed the design with the founder: https://news.ycombinator.com/item?id=48900288#48901679 (I think he bailed the conversation
17.
by infogulch 1mo ago
I think I'd first seen the idea on HN, but I don't recognize your comment specifically.
18.
by infogulch 2mo ago
They get the money from the peer networks that are the source of the spammers. If they don't cough up then you void your peering agreement. If they try to route around through a different trusted peer then that peer takes the hit. If i
19.
by infogulch 2mo ago
Why? They get paid $10 every time spam is reported regardless of the outcome. Trained staff listening at 3x speed with AI assistance could classify 3 obvious spam calls in a minute. If you can't turn $1800/hr per employee into a p
20.
by infogulch 2mo ago
The best technical solution I've seen is a spam review fee that turns into a reward and charged to the next closest originating peer. For example say you get a spam call. You notify your service provider and pay $10 to have them review
21.
by infogulch 2mo ago
Merged, thank you Richard! :)
22.
by infogulch 2mo ago
Unfortunately their Zed agent is completely broken when running on Windows connected to WSL/SSH remote because the terminal tool checks for paths using the local Path style (windows) which breaks when applied to paths on the remote (li
23.
by infogulch 2mo ago
Probably needed an excuse to rotate our stock out.
24.
by infogulch 2mo ago
LLMs are very good at helping you manage DMARC/DNS related configuration, even as a non-expert. I used it to develop custom DMARC report processing app that: 1. sucks in reports sent to our dmarc inbox into a sqlite db, 2. displays the
25.
by infogulch 2mo ago
I put a path in my garden so I can continue to maintain the garden, and walk around and enjoy it. But oh no, my daisies are wilting because they're too close to the path. So I adjust the path, raise the bed a little bit, and add a bord
26.
by infogulch 2mo ago
"be fruitful" and "replenish the earth" is already aligned As for "subdue", ask anyone maintaining a large garden how much they have to work to keep it from reverting back to a wilderness. Gardening is choosing
27.
by infogulch 2mo ago
I guess another thing is once the related conversation is identified you can use some other mechanism to retrieve it all or in parts etc.
28.
by infogulch 2mo ago
So you draw some distillates out of the conversation, do a vector embedding of each, and insert the pairs into pg. Then retrieval is distill -> find top N nearby vectors. Indexing, etc. Seems like a smart way to organize long term memory
29.
by infogulch 2mo ago
Yes that would probably help!
30.
by infogulch 2mo ago
LLMs are far from great writers. They struggle to form long coherent sentences and lean on punctuation like emdash and semicolon to ensure grammatical correctness when splicing together short phrases. This makes me wonder if the reason why

More ›