@alexjc@creative.ai
@alexjc@creative.ai avatar

alexjc

@alexjc@creative.ai

Creative AI enthusiast, building tools and teams where humans > machines.

See also my community-building bots https://creative.ai/@arxiv and @genart!

This profile is from a federated server and may be incomplete. Browse more on the original instance.

alexjc, to random
@alexjc@creative.ai avatar

The rig that powers the @arxiv bots had some trouble earlier in the week, so can't resume operations without some additional work on my part... No ETA unfortunately.

(I think it'd be best to use an online API now instead of Galactica anyway. Perplexity's beta API is great!)

alexjc,
@alexjc@creative.ai avatar

All the topic-specific bots continue to Boost what others share across the Fediverse though!

See: @arxiv_lg @arxiv_ai @arxiv_ne @arxiv_cv @arxiv_cl @arxiv_ir @arxiv_hc @arxiv_sd @arxiv_ro @arxiv_gr @arxiv_ma

alexjc,
@alexjc@creative.ai avatar

Turns out the 4,2 update of Mastodon has closed up the streaming API end-point that the bots relied on to get their updates.

There are ways to access this with OAuth tokens, but unfortunately due to a bug it doesn't work for clients (only registered users).

Not sure how to fix this yet. If anyone is willing to spend 250-500 EUR on a bounty, there's a developer willing to fix the client authentication bug!

StuartGray, to ArtificialIntelligence

Anyone know what's up with the TimeStamps on HuggingFace lately?

On at least one users repo (TheBloke), his last updated timestamps keep getting mass updated, affecting all files in the repository.

It's just happened for the second time in 3-4 weeks, and it looks suspiciously like a mass data restoration, but with no official announcement.

His work rate is so prolific that it's almost useless to follow without being able to sort by recently updates.

alexjc,
@alexjc@creative.ai avatar

@StuartGray Any changes in the Git history itself?

eloquence, to random
@eloquence@social.coop avatar

@alexjc The @arxiv bot is very cool. How do you generate the summaries?

alexjc,
@alexjc@creative.ai avatar

@eloquence Galactica30b running in 8-bit mode. I also experimented with Galpaca but nothing conclusive...

alexjc,
@alexjc@creative.ai avatar

@eloquence I learn most with local LLMs but it takes GPU-time away from other things!

I don't have contact with Anthropic.

airpods69, to ai

Can someone point me to the server or people where I can find those into AI research (deep learning too)?

Also to cool things related to linux (like NixOS?)


alexjc,
@alexjc@creative.ai avatar

@joerocca @airpods69 I don't use a Chrome Extension, it's a hosted service that previously used the Twitter API. I don't know if they're paying for the API now or if they scrape Twitter externally, but it still seems to work.

alexjc, to ML
@alexjc@creative.ai avatar

RT @alexjc

Want to know why The Open Web is dead? Because efficient-infringers & anti-Copyright activists decided on new internet norms to suit , now everyone is paying the price.

Not too late to change sides to become a © Jedi, Andres, dark side won't miss you! https://twitter.com/technollama/status/1674896586277629957

🐦🔗: https://twitter.com/alexjc/status/1674905104758054913

alexjc, to random
@alexjc@creative.ai avatar

RT @goodside

i find it pretty funny we can teach grammar to a <10M model (vs. 1.5B for GPT-2) but not on any human text; we have to train a >10000x bigger model on all the world’s knowledge and ask it to pretend to be a 3-year-old first. https://twitter.com/eldanronen/status/1658321669407248387

🐦🔗: https://twitter.com/goodside/status/1675030073525641217

alexjc,
@alexjc@creative.ai avatar

@moirearty Hmm, noted. 🤔

alexjc, to random
@alexjc@creative.ai avatar

Do the spikes in the red line look bigger than the spikes in the green line to you?

Over the years, a lot of bugs and edge cases have been fixed with the NVIDIA stack and it seems to affect numerical stability. Can likely be fixed, but takes time.

RT @abhi_venigalla

And yes, you can switch back and forth between NVIDIA and AMD, even within a single training run.

It's Christmas in July!🎄

🐦🔗: https://twitter.com/abhi_venigalla/status/1674795331756908544

alexjc, to random
@alexjc@creative.ai avatar

RT @alexjc

The breakdown we're seeing is absolutely due to people not respecting Copyright!

There's no going to the way it was back unless we find ways for people to respect ©. Otherwise, people and companies won't make things public anymore — now in progress. https://twitter.com/matdryhurst/status/1674901031111589890

🐦🔗: https://twitter.com/alexjc/status/1674903973311008768

alexjc, to random
@alexjc@creative.ai avatar

RT @swyx

🆕 Essay: The Rise of the AI Engineer

https://www.latent.space/p/ai-engineer

Keeping up on AI is becoming a full time job.

Let's get together and define it.

🐦🔗: https://twitter.com/swyx/status/1674826723068903425

alexjc, to random
@alexjc@creative.ai avatar

RT @TitterDaily

NEWS: Twitter's web version no longer allows users to browse without logging in. All urls redirect to the signup page.

This is believed to be a measure to make it harder for scrapers to take Twitter's data, like ChatGPT's web browsing plugin has been doing.

🐦🔗: https://twitter.com/TitterDaily/status/1674765247255654400

alexjc,
@alexjc@creative.ai avatar

If you want to stop OpenAI from scraping Twitter, ask them for their User-Agent and crawler IP addresses, then setup robots.txt and throttle the IP addresses.

It's best practice, surely that can be enforced without taking down all public Tweets?

alexjc,
@alexjc@creative.ai avatar

I'm always doubtful of the motives behind Twitter's new features, because the chairman playing The Fool publicly I consider theater — a good way to misdirect the competition and hide other ulterior motives (likely don't benefit anyone).

alexjc,
@alexjc@creative.ai avatar

I do, however, foresee the need for micropayments on such transactions — because you cut out the frontend + ads and therefore the platform and/or creator needs to be paid in other ways.

Shame Twitter is cowardly backing down from that discussion!

alexjc,
@alexjc@creative.ai avatar

News sites will die to LLMs unless we figure out those micropayments or customer subscriptions that include APIs.

alexjc,
@alexjc@creative.ai avatar

Context, Elon finding out ChatGPT can scrape threads. 👇

Apparently it's a temporary change due to scraping; would be nice to see Twitter use legal action to force known scrapers to use User-Agents, respect robots.txt, and data-mining best-practices. https://twitter.com/WholeMarsBlog/status/1668810759235854336

alexjc,
@alexjc@creative.ai avatar

The breakdown we're seeing is absolutely due to people not respecting Copyright!

There's no going to the way it was back unless we find ways for people to respect ©. Otherwise, people and companies won't make things public anymore — now in progress. https://twitter.com/matdryhurst/status/1674901031111589890

alexjc,
@alexjc@creative.ai avatar

Want to know why The Open Web is dead? Because efficient-infringers & anti-Copyright activists decided on new internet norms to suit , now everyone is paying the price.

Not too late to change sides to become a © Jedi, Andres, dark side won't miss you! https://twitter.com/technollama/status/1674896586277629957

alexjc,
@alexjc@creative.ai avatar

Berne Convention specifically says, you can't prejudice the legitimate interests of authors when you claim Fair Use...

Everyone forgot that (e.g. bots scraping content free without ads), now the web is dying.

alexjc,
@alexjc@creative.ai avatar

📊 Pick your future, chose wisely!

freakazoid, to random
@freakazoid@retro.social avatar

Pix2pix works by training a "discriminator" network to distinguish the output of a "generator" network, while simultaneously training the generator to fool the discriminator. Unlike with "unconditional" GANs, the discriminator gets to use the input image as well as the output of the discriminator, otherwise the generator could produce something that looked like any of the training outputs for any training input.

You may wonder how you start training a setup like this, since the generator will initially just output noise. The answer is that the generator also gets trained to minimize the L1 norm, i.e. the mean absolute difference between its output and the training image. By itself the L1 norm produces blurry results, which is the whole reason for the discriminator network.

As for how to update the weights, PyTorch or any machine learning framework will automatically produce gradients for you, since that's pretty much the point of those frameworks. So we know how to adjust the generator's output to get a better score from the discriminator, and from that we can compute an adjustment to the generator's weights. And we know how to adjust the discriminator's weights to better discriminate between the "real" and "fake" output images. It's not a game, so both sides get to "cheat".

https://arxiv.org/abs/1611.07004

alexjc,
@alexjc@creative.ai avatar

@freakazoid The bot used to monitor public streams of AI/ML related servers. If anyone on those servers follows you, your post could have ended up there, and then it spotted arxiv.org in the string, checked the paper by category.

The public streams don't work anymore though, it's WIP to fix that...

alexjc,
@alexjc@creative.ai avatar

@freakazoid "It's old." is not a strong criteria for the code that selects highlights (logarithmic decay of time). It's more interested in novelty or other factors...

I wrote the code for that, but it still looks random to me, it was one of 16k posts in the backlog that could have been selected.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • provamag3
  • thenastyranch
  • magazineikmin
  • ngwrru68w68
  • mdbf
  • rosin
  • Youngstown
  • vwfavf
  • slotface
  • modclub
  • khanakhh
  • cubers
  • kavyap
  • DreamBathrooms
  • cisconetworking
  • PowerRangers
  • everett
  • Durango
  • InstantRegret
  • osvaldo12
  • tester
  • normalnudes
  • tacticalgear
  • ethstaker
  • GTA5RPClips
  • anitta
  • Leos
  • megavids
  • All magazines
    •