dwheeler

18 ms·

241.
by dwheeler 3y ago
For context: The latest large language models (LLM) skip doing some complex things that previous good models did, and focus on attention. See: "Attention is all you need" https://arxiv.org/abs/1706.03762
242.
by dwheeler 3y ago
Check out: https://spin.atomicobject.com/2016/08/26/makefile-c-projects... https://make.mad-scientist.net/papers/advanced-auto-dependen...
243.
by dwheeler 3y ago
I disagree: it's about humility. We all make mistakes. Having a review process means we are humble enough to admit it.
244.
by dwheeler 3y ago
It's also a delightful answer when you're "in a pub having an argument", if no one thought of that as a possible answer. Something that makes you smile is a good thing.
245.
by dwheeler 3y ago
I used GNU make to manage a data pipeline with a large number of (mostly Python) scripts written by a team. It had to process lots of data using many steps. Make worked very well for this use case. It easily auto-generated dependencies, r
246.
by dwheeler 3y ago
Has anyone investigated to see if OpenCyc can be converted to natural language (presumably English) and then injested into this? Cyc made an attempt years ago to "encode common sense" and a subset called OpenCyc was released. That
247.
by dwheeler 3y ago
Ok, all evidence I can find is that the LLaMA model is not open source, and the article is just wrong. Software != model. If someone has evidence otherwise, please post.
248.
by dwheeler 4y ago
That is for code to download the model, not for the model itself.
249.
by dwheeler 4y ago
> Thanks to the open-source efforts like LLaMA ... Is the LLaMA model really open source now? Last I checked it was only licensed for non-commercial use, which isn't open source software at least. Have they changed the license? Are
250.
by dwheeler 4y ago
My understanding is that there are big differences. Manning was an Army Private First Class, that is, enlisted military, who had specifically signed agreements to not disclose classified information. It's not exactly the same thing, bu
251.
by dwheeler 4y ago
LastPass' entire business model was about protecting passwords, and passwords still got leaked. Most prople want security, not "ability to sue" which is not at all the same thing.
252.
by dwheeler 4y ago
GNU make at least can do that.
253.
by dwheeler 4y ago
> And it all syncs across your devices, for free?! Really? My Linux devices? Android? Windows? I don't think so. I recommend considering one of the most important features of a password manager is that it doesn't force you to
254.
by dwheeler 4y ago
I have never heard of this game, but it reminds me of the logic dice game WFF 'N PROOF ... more info here: https://en.wikipedia.org/wiki/WFF_%27N_PROOF I did play WFF N' PROOF, and it was instructive.
255.
by dwheeler 4y ago
6502 also had some "little things". E.g., it's jump indirect didn't work across page boundaries: http://www.6502.org/tutorials/6502opcodes.html#JMP
256.
by dwheeler 4y ago
He may have "made it up" to improve sales, but from a certain viewpoint it's correct. If decide to measure the "value" of a network based on the number of node connections, then the number of connections for n nodes
257.
by dwheeler 4y ago
Yes, shebang.
258.
by dwheeler 4y ago
That's not just BNF in its usual sense though.
259.
by dwheeler 4y ago
Bash is very widely available. If you use bashisms like arrays, just use bash as the first line.
260.
by dwheeler 4y ago
14977 is a disaster, do NOT use it. See: https://dwheeler.com/essays/dont-use-iso-14977-ebnf.html IETF's RFC works, but it has its own problems. In particular, most people use regexes constantly, and that RFC is u
261.
by dwheeler 4y ago
Don't use the old ISO 14977:1996 specification for BNF. For a rationale, see: https://dwheeler.com/essays/dont-use-iso-14977-ebnf.html Quick summary: 1. It is unable to indicate International/Unicode characte
262.
by dwheeler 4y ago
Case is better at pure string comparisons, but I believe the -x example is wrong. Test is PRIMARILY influenced by the argument count, so [ "$x" = "value" ] always does a string comparison for equality, never anything els
263.
by dwheeler 4y ago
I mean for using POSIX standard (Bourne) shell mechanisms. If you use nonstandard mechanisms, then whatever.
264.
by dwheeler 4y ago
You don't need the x-hack for strings. In shell, a widely-recommended form for ANY expansion of a variable uses quoting unless you have special requirements. E.g., normally don't use $foo, use "$foo". Tools like shellc
265.
by dwheeler 4y ago
No need for that. In shell the preferred form for "and" and "or" is outside the test, and you always quote anything with a variable reference, like this: if [ "$x" = "foo" ] && [ -z &quo
266.
by dwheeler 4y ago
The nebullvm projrct and specifically chatllama, by themselves, are clearly OSS. They're licensed under Apache-2.0: https://github.com/nebuly-ai/nebullvm/tree/main/apps/acceler... However, many
267.
by dwheeler 4y ago
Testing is important, because there are many ways things can go wrong. Automated testing is critically important, because "manual testing" is often another phrase for "testing we plan to do but won't do often if ever
268.
by dwheeler 4y ago
This article really needs a "(2018)" marker. This article predates GPT-3 and GPT-2, it even predates the essay "The Bitter Lesson" < http://www.incompleteideas.net/IncIdeas/BitterLesson.html >
269.
by dwheeler 4y ago
> there needs to be a way to eventually break backwards compatibility and remove the cruft. Actually, no. There's always a cost/benefit ratio. You can easily create a beautiful, elegant system repateadly forcing backwards incom
270.
by dwheeler 4y ago
> POSIX won because it was first. Simple inertia keeps it there. Not true. OS/360 was released years before the first Unix (OS/360 in 1966; Unix development started in 1969). Multics' development also preceded Unix (indeed

More ›