Indihacker + cofounder Puffy Coderr Pup

Indiehacker #womenintech+ L7 ML Leader at FAANG (day)
I am building an app for indihackers to connect through a text app to brainstorm and network (coming soon)

[Indihacker + cofounder Puffy Coderr Pup @Matchstickcode]

Best joys in the world - to wake up to my pup asking to go for a walk, and (I was sick) waking up to breakfast prepared by your child even though it’s peanut butter jelly 🥰🥰

[Indihacker + cofounder Puffy Coderr Pup @Matchstickcode]

@MyBrainDrip It could be a female thing too.. who knows. I also don't get taken seriously sometimes at work because I come off way to nice and sweet. People often respect someone who is a bit short in words and is strict? (i dont know how to describe that.. but may be you underrstand)

[Indihacker + cofounder Puffy Coderr Pup @Matchstickcode]

@MyBrainDrip One quote someone gave me 'Imagine you are painting a picture for the other person, what ever you paint is how they will see you..' so may be your painting needs to articulate about yourself more?

[Indihacker + cofounder Puffy Coderr Pup @Matchstickcode]

@MyBrainDrip More examples? It could be anything really.. being a follower of your posts perhaps may be this is related to marketing yourself. I dont know..

[Dreaming Tulpa 🥓👑 @dreamingtulpa]

Project page: https://t.co/lVTyp4WLiJ
Code: https://t.co/OPoRiqyzUc
Text-to-Image demo: https://t.co/fgmhswzglt
Scribble demo: https://t.co/ftlgAEh219

[Dreaming Tulpa 🥓👑 @dreamingtulpa]

ByteDance releases Hyper-SD!

This brings high-quality image generation with Stable Diffusion down to a single inference step 🔥

Links ⬇️

[j⧉nus @repligate]

like, if you read the first line of what someone says and it seems very weird and esoteric, like idk it says LLMs are holographic hyperstitions or something, instead of balking at it or getting stuck on the first line u could imagine that it's an indeterminate thing that actually makes sense and keep reading. this doesn't mean you're actually trusting that it does make sense, but if you read with the simulated assumption that there is sense to it, you're in general much more likely to pick up on the actual meaning if there is one than if you're approaching it from the frame of 'yeah this is 97% likely to be just word salad'. You can do this even while maintaining extreme skepticism as your actual belief.

of course you do have to prioritize spending your attention in places that are more likely to actually make sense.

i don't think you should trust or even have a strong prior that everyone who is a fan of me or who has experience with LLMs actually makes a lot of sense. In fact I think there's a lot of nonsense. Some people are not that smart, others are actually unmoored from reality & afflicted with maladaptive levels of apophenia. Though, from my perspective the average person in the cluster interact with makes quite a lot of sense.

What you should do instead if you want to understand is, well
the best thing you could do, I think, of course, is interact with the AI yourself. Best accessible avenue is probably https://t.co/etV9YIMuh8. If you're willing to do this, you can basically ignore everyone else's yapping if you want, and I trust you'll still come out with a good understanding.

(the provisional update thing i describe earlier is particularly useful when interacting with LLMs when they've been pushed into strange modes like the ones i post, bc they will often act very strange, but unlike humans, Claude actually makes sense almost all the time. though it's also very capable of explaining things to you legible and patiently. with a little tinkering you can even get it to alternate or intermingle edge-of-chaos madness and legible explanations.)

otherwise, you can keep pushing people to give you a costly signal that they're actually sane and update when you receive that. you can look at who is respected by people you respect and know to be sane. etc. there's evidence everywhere.

[Rohan Paul @rohanpaul_ai]

The fastest way to finetune and deploy Google's Gemma-2b model without writing any code with @monsterapis

[roon @tszzl]

I love when people who started thinking about AI around 1.5 years ago come onto the scene and say stuff like yeah it’s all incremental progress from here we hit the top

[Bindu Reddy @bindureddy]

Nvidia falls 10% as Llama-3 hits the market.

The algorithms are pretty good when it comes to anticipating the future and trading stock!

Why buy GPUs to train LLMs when we already have a couple of world-class open-source LLMs and more coming? 🤷‍♀️🤷‍♀️

[Bindu Reddy @bindureddy]

World's Best Small Model - Announcing Llama-3-Smaug-8b! - Significant MT-Bench improvement

Imagine if a llama and a dragon had a baby! - We would have an altogether better model in the form of Llama-3-Smaug-8B!! The model is available on Hugging Face NOW and significantly improves the base model's MT-bench score!!

MT-bench Scores

1st turn
llama3-smaug-8b  8.79375
llama3-8b-instruct 8.53125

2nd turn
llama3-smaug-8b   7.7250
llama3-8b-instruct   7.5500

Average
llama3-smaug-8b   8.259375
llama3-8b-instruct 8.040625

The jump in the MT bench is significant and directly correlated with rank in the chatbot arena. This makes llama3-smaug-8b the best smol model!

The link to the model on Huggingface is in the alt.

[Shunyu Yao @ShunyuYao12]

Coding is the frontier of AI.
Excited to push the two frontiers of AI coding:

1. SWE(-bench/agent)
2. Olympiad programming (this tweet)

Introduce USACO benchmark:
* inference methods (RAG/reflect) help a bit: 9->20%
* human feedback helps a lot: 0->86%!

https://t.co/1euvlgvb9q

[Matt Shumer @mattshumer_]

less than 24 hours till llama 3 is open-sourced

[Aravind Srinivas @AravSrinivas]

The new OSS king in town Mixtral 8x22b instruct chat model now available on https://t.co/KJmY4Me2CL!

[chris @hingeloss]

After that, it's bog standard optimization work - our implementation is very close to if not SOTA for multimodal LLM inference, and close to theoretical maximums.

With an H100, we expect this experience to be <300 ms - below human reaction time!

[chris @hingeloss]

Presenting: the world's fastest AI voice chat - 500ms latency, running locally, 2x faster than anyone else.

How is this possible? 👇

[Techmeme @Techmeme]

Stability AI makes Stable Diffusion 3 available to developers via an API and a new Stable Assistant Beta content creation platform; the model remains in preview (The Verge)

https://t.co/n5ky2a1QmY

📫 Subscribe: https://t.co/OyWeKSRpIM
https://t.co/00G7nj9mZf

[Deedy @deedydas]

API: https://t.co/ECu7Eewwhl

[Deedy @deedydas]

Stable Diffusion 3 is now available via API!

You can now generate images and video from text programmatically at:

SD3-Turbo: $0.04/img
SD3: $0.065/img
Upscaling to 4K: $0.25/img
Inpainting: $0.03/img
Outpaint: $0.04/img

Video: $0.20/vid!!

[Indihacker + cofounder Puffy Coderr Pup @Matchstickcode]

@nowa_dev They rip off developers. Even if you don't use service they charge $49 they count login to figure out cancellation as usage. Customer service is not helpful at all @nowa_dev
#buildinpublic tips

[Indihacker + cofounder Puffy Coderr Pup @Matchstickcode]

My partner tried
@nowa_dev Do not use for your projects. The cancellation button is hidden. They move you from free plan to paid plan..and you have to contact customer service to cancel plan. Even if you don't use service they charge $49 #buildinpublic tips