Mark Edmondson

...as I was saying bsky.app/profile/nrem...

Still a way to go before this is true but heading in that direction.

Ambition unbounded also means I’m looking to implement a new type of programming cache for AILANG based on embeddings so multi-agents don’t talk to one another, they share thinking space github.com/sunholo-data...

GitHub - sunholo-data/stapledons_voyage Contribute to sunholo-data/stapledons_voyage development by creating an account on GitHub.

AILANG is now in a good enough shape to start trying it out on a project, so I’m gonna use it to make a hard sci-fi philosophy game based on Starmaker by Olaf Stapledon github.com/sunholo-data...

Cosmic Paradox Reveals the Awful Consequence of an Observer-Free Universe | Quanta Magazine Encouraged by successes in understanding black holes, theoretical physicists are applying what they’ve learned to whole universes. What they’re finding has them questioning fundamental assumptions abo...

Observers are the boundary the create complexity? www.quantamagazine.org/cosmic-parad...

AILANG Documentation | AILANG Documentation AI-first programming language for AI-assisted development

Having fun making the most complex project I could think of with AI: creating a new compiled language for AI to use sunholo-data.github.io/ailang/ - teaching a lot about how to create complex systems keep on track with lots of self evals and iterative improvement

I’ll give him a go next then. Yes the anthill is self conscious in this story, if you know how to converse with it is another question though

I’m a little more believer of consciousness arriving from inanimate matter after the entertaining chapter in Gödel, Escher, Bach I’m just reading that gives a convincing argument from an Anteater that anthills can be conscious via emergent channels and symbols made up of unaware ants.

And certainly not borrow. That’s the underlying assumption why I think taxes will have to go up.

Ah I didn’t mean that - I don’t think the UK will raise taxes for infrastructure and get to Danish level of services - my prophecy is the UK Gov will be forced to raise taxes to pay debt and prop up deteriorating services, never being able to raise taxes enough for infrastructure investments.

It pretty much replaces pubs for friend maintenance so you need to get involved to be part of the community for otherwise taciturn Danes but it seems a lot healthier and it’s part of the fabric of society. No jobs paid but it enables volunteers

Exists in Denmark: if you can get 5 or more people to commit around an activity (sport, games, debating etc) then you can apply for a government grant to help cover expenses of building etc.

I predict that the UK tax burden for the middle class will be greater than Denmark’s within 5 years, and it will still leave the UK unable to invest in infrastructure

The bitter lesson when building AI applications is a negotiation between what people need right now, what the models can do via prompts, what guardrails need to be AI coded and what you could just wait 6 months for and will be possible as progress continues. It requires flexibility.

Thanks, gave it a retweet. I’m not building the models but am building the infrastructure and applications using them. It’s great fun

Would love to get someone who is working in machine learning/deep learning, or ideally, working on building LLMs to talk on the Search with Candour podcast with @jlwchambers.bsky.social. Let's give the actual experts the mic for a bit! 🎙️

If you know anyone, please tag them! ⤵️

My favourite Neal book

All of the winners are good, I'm always impressed by what people build with Quarto.

But let me highlight how cleanly mschnetzer.github.io/scrollytell_... combines scrolly-telling and regular layouts.

I didn't think "scrolly-telling is underused"; this submission moved my opinion in that direction.

AIdar (n): The uncanny ability to sense AI-generated content in the wild - for spotting those uncanny valley stock photos, suspiciously eloquent LinkedIn comments, and essays that somehow use 'delve' three times in one paragraph

This was a very enjoyable collaboration with @3blue1brown.bsky.social . Tanya Klowden and I are continuing to work on developing this material into a popular science book; we will share our progress on our Instagram at www.instagram.com/cosmic_dista...

The cosmic distance ladder with Terence Tao (part 2) YouTube video by 3Blue1Brown

I name my cats after the stars of this story youtu.be/hFMaT9oRbs4?...

How Core Git Developers Configure Git What `git config` settings should be defaults by now? Here are some settings that even the core developers change.

Fun new blog post today about some git config values that should probably be the defaults.

blog.gitbutler.com/how-git-core...

We must tolerate all but the intolerant is being tested to the limit now, since the intolerant are now accusing the tolerant of being intolerant, it’s easy to forget that the evidence of intolerance from tolerant people is being fabricated by the actual intolerants to create false equivalence

Marketing is often lies, to cover the weaknesses of the product. Classic example: Taste the difference (there is no discernible difference). If then I want to market Antisocial Lies, what could I call it? It’s going to get a lot harder when we have persuasive AI models writing the copy.

A much more likely target for being annexed into USA is the UK. It would solve a lot of Brexit issues, and that sense of superiority would be back in place to the relief of some.

If we end up creating superintelligence what I find scary is we will likely have people outsource their own personal responsibility to it, basically I expect new religions and cults. Already happening: “Anthropic-bros” school kids asking Claude what to do for every decision; talk to a girl etc.

Deepseek R1 671b Running LOCAL AI LLM is a ChatGPT Killer! YouTube video by Digital Spaceport

Here’s a video on what you need to do if you want to run DeepSeek R1 (the full 671b) locally. youtu.be/yFKOOK6qqT8?... - an hour to answer a question when 6tokens/sec

But $1000s of dollars per question spent on questions like “should I merge with this company?”, “what strategy should I take negotiating with this country”, “where can the best gains to AI be found given our current codebase” available in huge data centers. Exponentials not understood by public.

The economics of investing in GPUs are more assured, not less. Inference time compute (that DeepSeek took advantage of copying o1) is going to demand more capex not less. If there is no upper cap to demand for intelligence, which I can’t see at all. Open source local on your device for simple tasks.

It’s takes like these which is why I wanted to work with GenAI so as to be informed as we go through this period. I have opposite conclusions from the same event: the DeepSeek model is exactly on trend to where we thought open source should be 12 months ago. It’s going to mean accelerated progress.