Using coding agents to completely rebuild the Adobe stack and open source it is exactly what I was saying that this technology is going to be more liberatory than people make it out to be.

https://getartcraft.com/apps

#AI #LLM #Coding

The Transformer, and all other insane tricks to make attention efficient, are actually some of the coolest pieces of tech ever invented

#ai #transformer #attention #ml

I think about this a lot whenever I get reply guys #AI #SWE

LLMs don't need to self-replicate and take over all hyperscalers/neoclouds to be a genuine danger to society, hacking just 1% of the industrial control hardware that's connected to the internet is way more than enough to cause serious casualties. That 1% capability is demonstrably here, today, even with open weight models.

#AI #AISafety

I want to hear the arguments people have for thinking we can put AI back in the bottle.

I'm so tired of the progressive sphere not wanting to engage with something that's clearly here, remoulding our societies, and is asymmetric in nature (i.e. 90% of the world not using AI wouldn't matter, because the money is in enterprise deals not consumer side).

The conversation is captured by accelerationists, and outside the FALC-Left, I haven't heard a peep about how we handle this new world in a pro-humanist way.

#AI #Society #Leftism #FALC

Helix is amazing omg why have I been torturing myself with vim motions when there are better alternatives?

I never thought I could move off neovim, but I think Helix might be pulling me away.

#nvim #vim #helix #linux

China is Writing a National Rule that says AI Chatbots must let you talk to a human

Frustrated by chatbots that answer everything but the question, China is introducing a national standard on September 1 requiring companies to provide a clearly marked, functioning route to a human agent.

The standard, the country's first focused on the coordination of artificial and human service, requires that the manual entrance be easy to find, that complex disputes be promptly handed to a real agent, and that cases touching life or property safety be transferred automatically.

https://www.trendingonweibo.com/hotwords/ai-customer-service-human-escape-hatch/

#NoAI #China #TechPolicy

Abstaining from AI use will earn you a pat on the back... and that's about it. Good job, buddy.

- Frontier labs aren't demand-financed by marginal chat users, they're financed by capital markets betting on enterprise substitution (Anthropic had its first "positive" quarter when switching Enterprise plans to API rates)
- this card probably indicts data centre and training capex, and local models are the only true threat to data centre capex, simply due to token economics of these models
- Coding agents, imo, are atm a golden chance to massively provide open solutions to previously proprietary tech (think phone modem binary blob reverse engineering, ...).

@jwz https://mastodon.social/@jwz/117106720849031751

Can't wait to be hunted by the Palantir FPV Drone that was battle tested in Bakhmut because I escaped my data labelling pod to search for non-synthetic food in the wastelands. My body will be mulched up and used as biofuel for the GPT-7 data centre

Romanian conspiracy theorists believe it's likelier that Germany and Austria stole water from the Danube than that climate change caused historic droughts in Europe

#ClimateChange #Heatwave #EUPol

Morning, today we're looking at #MechanisticInterpretability , a subfield of AI Safety that attempts to understand the inner workings of artificial intelligence by analysing concrete structures, algorithms and circuits.

Why do we even need to do this? Because the hope of understanding neurons as being features died on polysemanticity -- models represent more features than dimensions by assigning them to an overcomplete set of non-orthogonal directions (i.e. you can't hope that concepts can be broken down into linear combinations of features).

This isn't a new field, people have been projecting intermediate GPT layers through the final layer activation and looking at the top-k most likely next tokens since GPT-2.

And that's how we arrive at the Natural Language Autoencoder -- an autoencoder over residual-stream activations where the bottleneck is natural language text instead of a sparse vector.

Nothing in the training objective requires the verbalisation bottleneck to be readable, faithful, or even semantically related to the intermediate layer being investigated, so while the training algorithm is typical of autoencoders, two things are distinct here:

1) both parts of the autoencoder (here the activation-verbaliser, AV, and the activation-reconstructor, AR) undergo special SFT to ensure the verbaliser and reconstructor can generate and read natural language from intermediate layer activations
2) a KL-divergence penalty is baked into the objective to make sure that, during joint training, the NLA doesn't diverge from the SFT version

A side-product of this training is that we can take the current verbalisation and "desired" verbalisation and compute steering vectors from their difference.

Surprisingly, Anthropic didn't find any evidence that the NLA was engaging in steganography to smuggle information to avoid human detection, but it did find that NLAs tend to confabulate a lot.

#AIResearch #Transformers #TransformerCircuits #Anthropic #NLA

Spain reports that 50,000 asylum seekers illegally crossed the border into Ceuta yesterday.

Ceuta's population is 83,000...

Even if you have the noblest of hearts, you have to recognise this isn't sustainable, and that in the race to take care of everyone, the systems of care we currently have will end up serving no one due to sheer overwhelm

#Spain #Ceuta #EUPol

The levels of self-absorption within all ranks of EU institutions can be simply summed up using Juncker's curse:

"We all know what to do, but we don’t know how to get re-elected once we have done it."

Members of the EU Council, Commission and Parliament all believe that they're moral reformists that are selflessly saving the EU from disaster, when instead are taking corporate bait hook, line, and sinker.

Metsola's actions this week are a testament to this.

#EU #EUPol #ChatControl #Privacy #DigitalRights

Surprise surprise, the state-adjacent, industry-funded quasi-regulator Internet Watch Foundation that pressured the European Parliament to fast-track Chat Control 1.0 through is also campaigning for the end of E2EE, without directly calling for it. Under the guise of "upload prevention", the IWF campaigns for on device scanning and on-device censorship

As always, this "grassroots" movement is funded and supported by Meta, Microsoft, OpenAI,

https://www.iwf.org.uk/policy-work/preventing-the-upload-of-child-sexual-abuse-material-in-end-to-end-encrypted-e2ee-environments/

#ChatControl #EU #Privacy #DigitalRights

I realised my inability to quit using coding agents during my self-imposed "AI Lent" wasn't due to willpower, but workload. Being overwhelmed with tasks made me pull the slot machine arm, hoping to land on a generation that will make the painful problem go away.

Tackling which work I'm agreeing to do, and which work I'm refusing was the key to letting myself slow down and enjoy coding without assistance again.

#AILent #Claude #Programming #NoAI

There's an insane number of IMSI catchers in Taipei. I get GrapheneOS pings almost daily.

Anyone know what this is about? News mentions SIM card fraud.

#Stingray #IMSI #privacy #cybersec #infosec

Mastodon doesn't suffer from a purity testing culture /s

@aburka https://hachyderm.io/@aburka/116491352327476996

Some shots of my new bike, for posterity of course

@roadbike @bikenite @fedibiker

#cycling #BikeTooter #RoadBike #單車 #公路車

AI is replacing workers at faster and faster rates, but by proxy, not causally.

> Meta to cut 10% of jobs, or 8,000 employees, report says

Data centre build out, training, and inference capex is sucking the money out of every nook and cranny these companies still hide their piles of cash made during the Covid years.

Naturally, some departments will be cut so that capital can be rerouted to AI spending

https://techcrunch.com/2026/04/23/meta-job-cuts-10-percent-8000-employees/

#AI #Meta #Labour #WorkersRights #Layoffs

The Left has an AI problem.

The biggest misconception is that LLMs and AI in itself carries a political ideology (the way _some_ people here characterise it).

That couldn't be further from the truth with any technology.

Keynes' prediction of 15 hour weeks by 2030 would be true given technological advancements and productivity improvements over the last 100 years, but because capitalism allocates resources to maximise profit, not for fulfilling needs, we're stuck with the 40 hour work week, wage stagnation, unemployment as tool for creation of Marx's reserve army of labour.

AI is also the impetus the Left needs to finally reach the white collar worker that is primed for radicalization due to job loss via automation.

If you want to win a war, denouncing your enemy isn't enough.

https://www.youtube.com/watch?v=Z75hlhgchQ4

#Leftism #AI #Socialism #LateStageCapitalism