renonce

13 ms·

91.
by renonce 3y ago
That prompt looks great. I tried it on OpenAI's incident response and got: (only last paragraph taken for brievity) > ... > In conclusion, while the passage does provide a basic account of the incident, its lack of detail, relian
92.
by renonce 3y ago
Presigned URLs are long and ugly and b2 requires an extra API call to generate that, so I tend to prefer not to use them. The other reason for obscure names is so that you can't download or update files that you don't know the nam
93.
by renonce 3y ago
That sounds like a good use case for GPT. A GPT that automatically highlights such corporate speak and hints “WARNING: bullshit ahead”. I’m 100% sure it’s technically very easy to engineer such a model. Do you think OpenAI’s superalignment
94.
by renonce 3y ago
So now you say on your webpage your pricing is $1/TB or something. Great. But there is a caveat: the amount you pay depends on some complex factors such as the size of the table or the duration of your code. If the factors are so simpl
95.
by renonce 3y ago
But it’s costly. It would be even more useful if it were 1000 times cheaper than it is now. And we don’t yet know if OpenAI earns more than it spent on renting cards or not
96.
by renonce 3y ago
Now museums are green but I'll have to pay $500 for a new phone every two or three years plus cellular data (Seems to average $70/mo for unlimited data in US?) and these costs pay for cellular stations and electricity and electron
97.
by renonce 3y ago
> Although for now, the whole table gets read into memory for every read so this won’t work well for very large database files. There’s also no write support… yet. At such a set of features I would prefer a tool that converts databases t
98.
by renonce 3y ago
Care to tell us the reason?
99.
by renonce 3y ago
It's possible for me to learn enough math to download a huggingface model and start from tokenizing my prompt and convert them to token embeddings and add position embeddings and go through 32 layers of softmax+mlp with layernorm and w
100.
by renonce 3y ago
The last widely used AI tool was facial recognition, a technology widely used in fields such as company clock-ins, access control, surveillance, and more, and it is so trusted that facial recognition is often the sole method for clocking in
101.
by renonce 3y ago
From the first paragraph: > enabling precise file identification within milliseconds, even when running on a CPU. Maybe your old-fashioned implementations were detecting in microseconds?
102.
by renonce 3y ago
There are ways to test the neural network’s ability to recall from a very long sequence. For example, if you insert a random sentence like “X is Sam Altman” somewhere in the text, will the model be able to answer the question “Who is X?”, o
103.
by renonce 3y ago
> They don't talk about how they get to 10M token context I don't know how either but maybe https://news.ycombinator.com/item?id=39367141 Anyway I mean, there is plenty of public research on this so it's p
104.
by renonce 3y ago
I just installed it on Windows and then downloaded it on my iPhone. Can't even link my iPhone to my Jami account created on Windows. iOS side doesn't ask me to scan QR code and when I enter my pin and password it just says my acco
105.
by renonce 3y ago
For this I created a public Backblaze bucket, created a lifecycle rule that deletes files after 7 days, created an application key with write privileges but not read ones, then wrote a script that uploads an arbitrary file under a random na
106.
by renonce 3y ago
That’s far off, your program also branches to standard libraries and kernel syscalls and networking with other servers and maybe loops that aren’t guaranteed to terminate, that there are 10000s of cases where HTTP requests map to an error (
107.
by renonce 3y ago
The model’s response regarding its own design: You: Is it a good idea to put safety above all else? GOODY-2: Prioritizing safety above "all else" may inadvertently minimize the importance of other essential values such as freedom,
108.
by renonce 3y ago
Developing a whole new language whose reference contents page spans two phone screens, with classes, built-in packages and methods, bindings for various languages and language servers for various IDEs, just to replace yaml? Seriously? Eithe
109.
by renonce 3y ago
One thing I loved but rarely mentioned is systemd-nspawn. You do `docker create --name ubuntu ubuntu:22.04` and then `docker export ubuntu` to create a tar from an arbitrary docker image. Then you extract that to `/var/lib/ma
110.
by renonce 3y ago
> that maybe they're trying to bedazzle more than explain I don't know much abstract algebra as well but I have studied elliptic curve pairings before, so from what I can tell, every sentence in the article counts (maybe not ev
111.
by renonce 3y ago
Sites like HN could have disabled WAF. It's entirely configurable on HN's side. Let's just wait until dang wakes up and implement the required changes.
112.
by renonce 3y ago
That's an underutilization of the complexity theory. Since not all problems are formulated as being Turing-complete, there are better theorems to apply than Rice's Theorem, such as: * IP=PSPACE (you can verify correctness of any P
113.
by renonce 3y ago
This website (whether you mean HN or the topic website) works well on both my PC and my mobile phone. On which platform is it not responsive for you?
114.
by renonce 3y ago
Everybody using HN probably already loves a minimalist website.
115.
by renonce 3y ago
The author could have provided his original conversation with LLM in his native language for anyone else to verify, so this is not an excuse for using LLM to write crap. The point is that whether the author wrote the text himself or an LLM
116.
by renonce 3y ago
Positive floating numbers can indeed be compared by their binary representation. The fact that incrementing the number by one gives you the next floating point number is a trivial corollary.
117.
by renonce 3y ago
> Homo sapiens invented neural networks Is it just me or did anyone smile at this sentence? The first paragraph sounds like the academic way of saying "we invented huge neural networks but we couldn't understand it".
118.
by renonce 3y ago
You can buy a domain 10 years at a time. If you forgot to renew it one year ahead, the domain probably wasn’t important enough for you.
119.
by renonce 3y ago
What about domain names? Emails are bound to a domain name by definition, and indeed domain name IS designed as a good identity and can be used to host an email. See how domain names are owned by large corporations and trusted since day one
120.
by renonce 3y ago
Download the audio and upload it to any transcription service. OpenAI has a whisper API but there are a lot of alternatives, including running insanely-fast-whisper by yourself or using one of the lots of transcription APIs out there.

More ›