oofbey

18 ms·

181.
by oofbey 8mo ago
This is a neat idea for a test. But the test is badly executed. A single comparison could just be a fluke. Compare it on a dozen tasks, trying each task a dozen times. Then you get data which is believable.
182.
by oofbey 8mo ago
Another common mistake today is to observe one LLM failing to do something in a single situation, and to generalize that observation to "LLM's are incapable of doing this thing." Or "they're not good at this kind o
183.
by oofbey 8mo ago
What have you managed to decompile? Did you do it with a coding agent?
184.
by oofbey 8mo ago
Claude code and cursor agent and all the coding agents can and do run MCP just fine. MCP is effectively just a prompt that says “if you want to convert a binary to hex call the ‘hexdump’ tool passing in the filename” and then a promise to t
185.
by oofbey 9mo ago
I’ve often thought one of Zuck’s superpowers is in finding ways to get smart and moral people to do truly evil things. Sometimes it’s mind games. Sometimes it’s careful layers of obfuscation. Here it might be: This analytics package is dyna
186.
by oofbey 9mo ago
I don’t know these OS’s well enough. Can you MitM the dynamic code loads by adding a CA to the OS’s trusted list? I’ve done this in Python apps because there’s only 2 or 3 places that it might check to verify a TLS cert.
187.
by oofbey 9mo ago
Agents meaning an AI iteratively trying different things to try to decompile the code. Presumably in some kind of guess and check loop. I don’t expect a typical LLM to be good at this on its first attempt. But I bet Cursor could make a good
188.
by oofbey 9mo ago
The entire MCP ecosystem disagrees with your assertion that “you can’t teach it new tools at inference.” Sorry you’re just wrong.
189.
by oofbey 9mo ago
What’s the state of the art of reverse engineering source code from binaries in the age of agentic coding? Seems like something agents should be pretty good at, but haven’t read anything about it.
190.
by oofbey 9mo ago
Such a refreshingly good attitude!
191.
by oofbey 9mo ago
How many people here rely on Google Docs? I have for many things for years. How many of you regularly back up your Google Docs? I have taken a Google Takeout a few times over the years. But no. Why? Because I have never heard of Google “lo
192.
by oofbey 9mo ago
If each client has its own database, does that mean you're automating the infra to turn on a new instance for each database, and then routing requests appropriately? Not too hard with k8, I suppose. Or if you add new clients infreque
193.
by oofbey 9mo ago
True. But “big beef” is complicated and difficult to make reliable. Horizontal scaling of unreliable servers is dirt simple to stay up through almost anything except sudden load spikes. And then it’s largely a matter of configuring your aut
194.
by oofbey 9mo ago
Huh. That sucks but also isn’t surprising. Seems like putting this in the cloud eliminates most of its benefits though. If you’re going to wait for that much latency you might as well use a “real” (traditional big complicated) rdbms.
195.
by oofbey 9mo ago
How do you scale your front end horizontally with a file based database? Do you put the files on a shared file system that the app layer all mounts and lock when writing? Or do you shard and route by user? Or do you build a big vertically s
196.
by oofbey 9mo ago
I completely agree. But “disclosing the use of AI” doesn’t solve that one bit.
197.
by oofbey 9mo ago
If there is more than one machine behind the NAT which one would it forward it to? This hypothetical simple NAT without firewall AFAICT doesn’t exist in reality, even if it exists in specs. I don’t see how it actually could.
198.
by oofbey 9mo ago
Integration tests are where it breaks down. Assuming Claude can just install whatever infrastructure it needs inside the VM only works for relatively simple projects. Claude probably isn’t going to install a Kafka queue in your VM. Or a k8
199.
by oofbey 9mo ago
Harsh sentiment. Pretty soon every knowledge worker will use AI every day. Should people disclose spellcheckers powered by AI? Disclosing is not useful. Being careful in how you use it and checking work is what matters.
200.
by oofbey 9mo ago
I understand ipv4 networks pretty well. And I would say that any device doing NAT is acting as a basic firewall. Do “true” firewalls do more? Sure. But saying NAT doesn’t provide security is flat out wrong.
201.
by oofbey 9mo ago
There are two spheres of influence you need to consider. The local machine/vm/container that the agent is running in. But also the effect the agent can have on the outside world - using auth tokens or ssh keys or apis that is has
202.
by oofbey 9mo ago
Nice. The big picture strategy is figuring out the right mix of deterministic and probabilistic programming elements and how to mix them. I see agents ignore clear instructions all the time thus we need to codify the important instructions
203.
by oofbey 9mo ago
Ironic that you nitpick the author’s word choice of “back pressure” and then completely misuse the term RL in your complaint.
204.
by oofbey 9mo ago
I like this idea but I can’t think of a concrete example to ground it. Can anybody share a real example?
205.
by oofbey 9mo ago
I’d like gastown more if it could run cursor-CLI instead of claude, and thus be able to choose models. Claude is okay. But these things certainly have personalities. I’m not sure which would be best for each role. But gastown’s different ac
206.
by oofbey 9mo ago
The big bet with this technique is in having a fixed (non learned) matrix which converts the tokens latent space to the linear attention space. So you can kinda cheat and say your model is small because a bunch of the smarts are in this fix
207.
by oofbey 9mo ago
Depending on how different the attention mechanism is, that might not work. If it’s just a faster / different way of finding the tokens to attend to, sure. But I get the sense the author is implying this method uses different semantics
208.
by oofbey 9mo ago
Services industry is gonna be tough in the age of AI agents.
209.
by oofbey 9mo ago
Please tell us more
210.
by oofbey 9mo ago
This is IMHO an impossible line to draw. If an established artist uses a GenAI music tool it would be accepted. If somebody unpublished does the same it wouldn’t. Assuming they can even tell the difference, which they can’t.

More ›