amluto

24 ms·

91.
by amluto 1mo ago
seL4 is willing to pay, but they're mostly targetting a different use case: running programs targeting seL4, which isn’t what typical agentic workflows want. You can, however, run Linux on seL4, and maybe agent sandboxes should start
92.
by amluto 1mo ago
The one time I tried asking Sol to use subagents for a small project, it took a surprisingly long time, used up the entire usage limit in one go, and basically failed the project. I’m pretty sure that plain Sol, serially, could have finishe
93.
by amluto 1mo ago
IMO the obvious answer is formally verified security. We can do this today for user mode, and we can mostly do it for ARM64 virtualization. It will be a while and would require substantial assistance from Intel or AMD to achieve it for x86
94.
by amluto 1mo ago
Surely yield is very much a function of the size of the device in question and its tolerance for defects. I don’t think it makes sense for a technology to have a theoretical maximum yield that’s a constant number unless it somehow destroys
95.
by amluto 2mo ago
I am not personally aware of any treatment for which there is particularly compelling evidence that it meaningfully slows progression even when started early. I wish there was. It’s also not clear that one can measure progression well enou
96.
by amluto 2mo ago
This is almost entirely dominated by the read circuitry and the data path: it’s still taking 1/6 of a second to read the whole chip, which means that the flash cells aren’t working hard at all. (And that pitting the full weights of a
97.
by amluto 2mo ago
Just to check: when were you growing up? In the CRT, it was often possible to see the physical pixels (i.e. the phosphors and the rows defined by the shadow mask), but there was not necessarily any guarantee that a logical pixel lined up w
98.
by amluto 2mo ago
My favorite thing about AWDL is that AirDrop barely even works between Apple devices that are right next to each other.
99.
by amluto 2mo ago
Ubiquiti has a half-hearted MLO implementation with AFAICT utterly silly restrictions on WPA mode.
100.
by amluto 2mo ago
I’m willing to believe that one could design a little circuit that multiplies a number stored in a floating-gate MOSFET by an analog input and another circuit that adds the result to an accumulator (in fact there seems to be some prior art
101.
by amluto 2mo ago
This, like almost every typing training system out there, suffers from my pet peeve: if you type an extra character or miss a character, then you end up with a series of “mistakes” instead of just one. Why don’t these programs use some form
102.
by amluto 2mo ago
Wow, that looks like a huge mess.
103.
by amluto 2mo ago
Right. But there is a lot of real-world usage of MCP-over-HTTP-over-the-Internet, and a lot of “harnesses” want to support that use case, so they’re stuck either implementing the HTTP-based protocol or using a shim.
104.
by amluto 2mo ago
Those are mostly at a different layer. You can speak Thrift or Avro or Protobuf over stdio or HTTP or TCP or carrier pigeon. gRPC spans layers, and it uses HTTP in a more intrusive way than even MCP does — it expects to own the entire URL s
105.
by amluto 2mo ago
I’m not really a fan. But if you’re building a protocol that needs to map to HTTP anyway, then maybe using the HTTP binding everywhere is not totally awful. In the flip side: I’m currently designing an AI-adjacent protocol, and it will be
106.
by amluto 2mo ago
The dimension section triggered my “vague math” spider sense: > During this zic-zac motion across the plane the vector still rotates along the plane in the except same ways as in the 2d case above. What is “the plane”? In 2D there is on
107.
by amluto 2mo ago
The prose on the page is very unclear. My best interpretation is that they want to continue supporting stdio but that they don’t want it to be its own special protocol. The obvious way to do that would be to speak ordinary HTTP (version 1.1
108.
by amluto 2mo ago
> Field testing is a real thing in literally all industries. I’m fairly confident that, if a company that makes door locks want to field test their locks, they test the lock and maybe the door. For some reason the software industry likes
109.
by amluto 2mo ago
Suppose an automaker creates a BankRobberGym and carefully trains the car to autonomously rob simulated banks because they think someone will pay them to use the car to legally test bank security, but they end up, predictably, training the
110.
by amluto 2mo ago
In my experience, a large fraction of flights have very stinky air during a substantial portion of the time from when the engines start until takeoff. Whether this is “gym sock” or something else has never been entirely clear to me.
111.
by amluto 2mo ago
Wow, that whole thread is borderline incoherent, presumably generated by an AI without adequate oversight. Here are the docs: https://developers.openai.com/api/docs/guides/prompt-caching... The thread has lit
112.
by amluto 2mo ago
I have very mixed opinions about the custom syntax. IMO the correct asm syntax, with very few exceptions, is the one in the manual . This is why Intel syntax is right and AT&T syntax is wrong: the ISA comes from Intel, the docs are fr
113.
by amluto 2mo ago
Since when? https://learn.chatgpt.com/docs/codex/cli https://github.com/openai/codex Looks like it's still a more-or-less-open-source product called Codex CLI. FWIW, there have been issu
114.
by amluto 2mo ago
> Cargo is a cross-platform tool, so when it ships a sandboxing solution it will need to be a cross-platform solution This seems like an excuse, not an actual objection. Linux can do seccomp or Landlock or gVisor or a combination. Seccom
115.
by amluto 2mo ago
If you mean chatgpt.com, try with a real “agent” harness. The exact same model on chatgpt.com frequently performs much worse than it does with a proper harness, not to mention that the chatgpt.com frontend itself is quite bad. codex-rs, p
116.
by amluto 2mo ago
I can easily clear lots of low-priority engineering work quickly using AI. It takes a bit longer if I want to do a good job. But here’s the rub: this is, basically by definition, low-priority engineering work. Those fixed papercuts are
117.
by amluto 2mo ago
The app’s ability to do complex things like opening the frunk is extremely poor, to say the least.
118.
by amluto 2mo ago
As an easy start, how about letting build scripts read /usr, read and write a temporary build directory, have some /tmp scratch space, and be allowed to write its final output artifact. No network and otherwise isolated from the r
119.
by amluto 2mo ago
Substantially better link: https://www.merck.com/news/merck-and-moderna-announce-phase-... Still no actual Phase 3 data presented.
120.
by amluto 2mo ago
A cursory estimate courtesy of ChatGPT suggests that there is a grand total of one order of magnitude or less of power efficiency improvement available compared to current Blackwell if the entire system’s power consumption outside the ALUs

More ›