cjbprime

31 ms·

391.
by cjbprime 3y ago
The difference is that no-one would load remote-user-supplied C++ code at runtime and expect it not to result in vulnerability, but people remotely source JS, or allow remote-user-supplied HTML to interact with their JS, and the stakes are
392.
by cjbprime 3y ago
As others are saying, this isn't fingerprinting for parents, or for attending school events -- it's for applicants to volunteer at the school.
393.
by cjbprime 3y ago
I use this too! I wish there was a better way to self-host keeping a vault up to date on the Android client, though.
394.
by cjbprime 3y ago
Nice! For a less bare bones OSS Kanban board, I like https://www.taiga.io/
395.
by cjbprime 3y ago
First, GPT-4 was trained on images and text, not just text. The images improved its text predictions. Because it has a world model, and they helped populate it. It outputs only text. But just like a Unix program writing text to stdout
396.
by cjbprime 3y ago
Nice article! I think you nailed it. The dismissal of AI reasoning and world modeling from famous technologists who should know better is such a disappointment. I have a general suggestion to avoid these word puzzles, because someone is j
397.
by cjbprime 3y ago
Your understanding of alignment is somewhat out of date. Training a model to produce human-valued responses and training a model not to decide to destroy all the humans are not separate problems. RLHF may actually be an excellent solution t
398.
by cjbprime 3y ago
I used gnus for email (and ERC for IRC) for a few years starting around 23 years ago. The showstopper was the lack of async I/O and multithreading in GNU emacs, so the more you do inside emacs, the more everything regularly freezes up
399.
by cjbprime 3y ago
> The latter requires complex cognitive faculties not present in a mathematical model of the English language. The training is on images and text, and the output is text. But nothing constrains the model in the middle of those two ends t
400.
by cjbprime 3y ago
> To oversimplify a bit, it's a next-word-in-the-sentence prediction engine. Turns out that when you ask LLMs to predict the next word in the sentence, and then train billions of parameters on billions of sentences, they realize the
401.
by cjbprime 3y ago
> Brooks: No, because it doesn’t have any underlying model of the world. I don't know whether to be more disappointed with the famous technologists who are apparently unable to think of questions to ask GPT-4 that require a world mo
402.
by cjbprime 3y ago
No, it's capable of reasoning, but it gained its knowledge of how the world works through analyzing things that were written or drawn by (mostly) humans.
403.
by cjbprime 3y ago
The previous model (Ampere Altra) is a Neoverse N1 which "includes two 256b SVE vector units". The new model discussed here (AmpereOne) is presumably their own core but I don't think they've released details.
404.
by cjbprime 3y ago
> presumably if you ask it to execute that same code for a input example you provide, it will. GPT-4 is run sandboxed, without the ability to execute code. The only way it can execute code is by attempting to simulate the algorithms, fro
405.
by cjbprime 3y ago
This is a misunderstanding of the researchers' claim, for which I blame them rather than you. You are using their study to claim that LLMs can not learn and reason about novel tasks. Their study doesn't make any claims about what
406.
by cjbprime 3y ago
This is a deeply incurious take. Yesterday I wrote a Python script, used dis.dis(code) to output its bytecode, and gave the bytecode to GPT-4. From the bytecode, it correctly decompiled the exact script I'd written (python bytecode in
407.
by cjbprime 3y ago
I guess that was interesting, but I'm mainly interested in what you felt about the output yourself. (I do wish people would be a little less willing to attempt to induce distress in LLMs for fun. It's not enjoyable to read for me.
408.
by cjbprime 3y ago
For what it's worth, I think it feels like you're yelling into a hurricane because you are kind of clearly incorrect. Can you think of a few questions (using unique nouns etc to avoid parroting) for which the answers require a wor
409.
by cjbprime 3y ago
Since we're setting up a database, has anyone seen a project to store stdout/stderr in tables too? Seems more useful in the GPT age, when it could provide an opportunity for conversations about the whole task you worked on. (As w
410.
by cjbprime 3y ago
Anyone know whether the plans to open-source the language are the "in a few weeks" type of plan or the "maybe never" type of plan?
411.
by cjbprime 3y ago
The two-channel DDR5 in desktops can't even do two channels very well -- if you try to put 64GB RAM in (two dual-rank 32GB DIMMs) then you lose around 50% of the bandwidth compared to a single rank DIMM (e.g. from 8GHz to 4GHz speeds,
412.
by cjbprime 3y ago
> I think that depends on how many millions we are talking about here, don't you? Not as much as it depends on how many millions somebody already has , no. Again, if you're sensitive to income loss, the answer would be to not
413.
by cjbprime 3y ago
> Individual humans can still easily beat it when it comes to knowledge of individual subjects. What does a phrase like "GPT-4 scores 90th percentile on the Uniform Bar Exam" mean to you, regarding whether humans can easily sur
414.
by cjbprime 3y ago
It would be a useful contribution to explain what you think the material differences are, rather than referencing them through innuendo, as if anyone knows what you mean.
415.
by cjbprime 3y ago
You sound like someone who's never asked GPT-4 to write a rap battle about $SUBJECT in the style of $CELEBRITY where every word starts with $LETTER..
416.
by cjbprime 3y ago
Do you really think it's incomprehensible that someone who is quitting so that they can talk freely would avoid signing documents that curtail their ability to talk freely ?
417.
by cjbprime 3y ago
> most bug bounties won't pay out XSS unless you can demonstrate that it actually leads to something. Because it almost always doesn't. This is weird. XSS usually leads to complete session takeover, and being able to perform ar
418.
by cjbprime 3y ago
Am not expert, but here's the understanding I've picked up: When you use JavaScript through Node, your JS gets the ability to open files and make network sockets and so on, because there are a set of APIs implemented by Node and e
419.
by cjbprime 3y ago
As others have said, it's an awful title. Could instead be something like "Is the emergence aspect of Emergent Abilities in Large Language Models a Mirage?". Like, there's supposed to be nothing academic researchers like
420.
by cjbprime 3y ago
But to be clear, you can think of no mechanism that would allow us to expose that incomprehension via communicating with it?

More ›