fswd

It's super slow. 1 token per second, if that. Half a word a second.

I am working on this and seeking a non-technical co-founder. I'm not sure it would make money however.

90% of frontend development is gathering the requirements. The other 90% is trying to get CSS to center a div.

definitely, I simply wasn't able to see the video until now.

the "bent propellor" could just be an optical illusion due to camera FPS vs reality.

What upsets me the most is that the execs gave themselves bonuses in to their personal FDIC insured accounts, essentially writing a bad check for free bailout money, after the bank went negative.

It's $3.80 for the full 1000 prompt tokens plus 32k token long context. But OpenAI doesn't have a monopoly on long context, RWKV is free and open source and has virtually an unlimited context (as long as you remind it every once and a while). However ChatGPT really cannot be matched at this point perhaps except LLaMa Alpaca 7B

EDIT: I accidentially typed Alpine, it's Alpaca. https://github.com/tatsu-lab/stanford_alpaca

They could probably brag about being FDIC insured, maybe make the font bold to restore their branding image.

Why a company with VC backing for something so simple? I am confused why somebody would think this is so important, they would need funding. Anybody with 1-2 years development experience should be able to hack something up in redis or even just crontab and flag files.

I use postgraphile graphile-worker https://github.com/graphile/worker for this.

For example, every month we roll over credits. For each user, when they signed up, 30 days from that, check. If they are available for roll over, reset and email. Then we have drip campaigns for alerts like running low on credit.

Also, if you upgraded your account, then pause payment, it uses a worker to schedule the date they are paid up to run the SQL to downgrade. With a simple API called 'addJob' that looks for a JavaScript file in a folder called task.

If you actually try and do this, the sales people will stop you due to some internal rule. No GPUs on free credit. Unless the situation has changed of course..

"model": "gpt-3.5-turbo",

turbo isn't listed in Playground, but if you invoke the example curl command (note: /v1/chat/completions) in your terminal, it works.

I've ran it on a AMD 3950 which I think is half the speed of a M1, and it's plenty fast. Note I am specifically talked about RWKV

What is pinecone and is there a link to a website?

I should clarify, I'm not suggesting you are censoring, or "site sponsored censorship", rather censorship from users via downvotes. They both accomplish the same thing I could understand why somebody might misunderstand me that you are censoring.

The Nordstream 2 thread was also interesting, with tons of users trying to appeal to you specifically to do site censor the story, or tone policing. I remember you stuck to your guns.

20B GPT-NeoX runs on a 3090 in 8 bit mode

Does anyone know if 16 will bring back JSON_TABLE from Postgres 15? They initially had it in the beta, then at the last moment, removed it.

It's even worse than this. Google's entire business was based on the browser search model. You want to know something, you opened your browser, then 100% of the time went to "google.com", searched, scroll,scroll,scroll,click and get a reward with your answer while seeing ads. Which pretty much describes a dopamine feedback loop.

The problem isn't that they don't have a ChatGPT ready (they do -- and it's probably better, I'm certain) .. but rather that, by even doing so (providing a chatgpt based search) they undermine their entire business foundation. Because now if you want to know something, you simply chat, and instantly presents a dialog with your answer. And with seemingly unlimited capabilities to write follow-up questions just like you would to a normal person.

How are you going to sell ads to that? Imagine talking to a chat bot that tries to sell you a new refrigerator when you're asking about something else, awkward? That's not going to happen.

That's why it's worse than a strategic mistake, it's an extensional threat-- they are damned to undermine the entire foundation in any direction they take.

what does the acronym AGW stand for? Can't figure out if you mean global warming or the story of Genesis.

For LLM, INT8 is old news but still exciting. FP8 would definitely be an improvement. However the new coolness is INT4.

> Excitingly, we manage to reach the INT4 weight quantization for GLM-130B while existing successes have thus far only come to the INT8 level. Memory-wise, by comparing to INT8, the INT4 version helps additionally save half of the required GPU memory to 70GB, thus allowing GLM130B inference on 4 × RTX 3090 Ti (24G) or 8 × RTX 2080 Ti (11G). Performance-wise, Table 2 left indicates that without post-training at all, the INT4-version GLM-130B experiences almost no performance degradation, thus maintaining the advantages over GPT-3 on common benchmarks.

Page 7 https://arxiv.org/pdf/2210.02414.pdf

Any suggestions for a good stack to develop extensions? I'm coming from nodejs/react, and would like to stay close to that. Is there an up-to-date stack I can clone, copy, and deploy with something like 'yarn run build' and start hacking?

You should definitely be able to read freely and fine at 45, and definitely see a doctor.

Something you can do in the in between time is order a full blood panel, kidney, liver, thyroid, hormones, etc, online or through your doctor's office. Then get a copy of this (you can often download a PDF from the lab company that does your panel, or doctor's office) and visit a natural path (ND). You might be able to find one through your school. Bring up what foods you eat, your lifestyle, exercise, sleep habits. Then, try and see if any adjustments work.

When you finally do get an appointment for a medical doctor (you said in 9 months) bring this information, so your doctor has something to go on. Your MD will probably just re-order the labs again, but now you will have some previous data, and some experience tweaking your lifestyle with what worked and didn't.

It's about 1/3 of a tablespoon. It's dangerous to the bottom line of drug manufactures trying to sell a $1000 injection that does the same thing.

It appears you can do the same thing with 3g of cinnamon

https://pubmed.ncbi.nlm.nih.gov/19158209/

Conclusions: Ingestion of 3 g cinnamon reduced postprandial serum insulin and increased GLP-1 concentrations without significantly affecting blood glucose, GIP, the ghrelin concentration, satiety, or GER in healthy subjects.

I use a monorepo (turbo) that has a directory: ./app/<docker workspace or app name>. In it I have a package.json that runs docker-compose up. I use regular .env files and keep a .env.sample in the repo. The docker-compose brings up the services and volume mounts the .env into it's local environment (same with /data for example postgres, or some other package).

I then disable iptables in /etc/docker/daemon.json: "iptables" : false to prevent docker from poking a hole in my firewall. (This should be on by default..) I also enable User name spaces with this: "userns-remap" : "default" so that the docker image runs isolated. I can disable this in my docker-compose with userns=host Actual instructions are slightly different, but: https://docs.docker.com/engine/security/userns-remap/

I also found I could disable networking all together on some services, and just expose a port. Some services don't need to connect to the net.

Then I run a cloudflare argo tunnel to the application. But you could also use a nginx reverse proxy, some other tunnel, or a one page express.js reverse proxy with pm2. There's a cloudflare argo tunnel docker-compose setup which allows you to type in your tunnel information in the config. I thought that was cool but I did not try it.

running all apps in something like a lite version of docker in user namespace mode would implement this

The language itself needs to evolve so that so many libraries are no longer needed. For example, in javascript if I can't call it via fetch, I don't implement it.

Someone did, now.

That isn't self hosted. I'm referring to running GPUs and models.

What is your use case? I could build this for you, email is in profile.

Yes this is true.