Gpt3.5 as used in the first commercially available chat gpt is believed to be hundreds of billions of parameters. There are now models I can run on my phone that feel like they have similar levels of capability.
Phones are never going to run the largest models locally because they just don't have the size, but we're seeing improvements in capability at small sizes over time that mean that you can run a model on your phone now that would have required hundreds of billions of parameters less than 6 years ago.
The G in GPT stands for Generalized. You don't need that for specialist models, so the size can be much smaller. Even coding models are quite general as they don't focus on a language or a domain. I imagine a model specifically for something like React could be very effective with a couple of billion parameters, especially if it was a distill of a more general model.
It seems that all of the comparisons with computational systems are either not really true (the supposed sharp distinction between hardware and software depends on whether you're considering the system as a software engineer, a firmware engineer or a hardware engineer. Computer systems are embodied just as much as any biological creatures) or contingent - if it were regarded as essential to consciousness that an organism have a source of true randomness for example, then we would simply add such a source to our systems (assuming consciousness was something we actually wanted them to have).
None of the descriptive points of what it means to be a biological organism really seem germaine to the core question of consciousness, which as far as I'm concerned is the inner experience.
Every month, the thesis seems more certain that a mathematical model will be able to produce output indistinguishable from the output produced by a conscious, biological creature.
Once that's accepted, then the only interesting questions that are left are by definition unobservable. Is there anything that it is like to be that mathematical model?
I was not particularly a fan of them - the plot seemed to find overly easy solutions to all the actual messiness that comes when dealing with others very unlike yourself, which given the rest of the stories, feels like it undercuts the entire point of them.
The Tchaikovsky novella I really like is Elder Race. Technology-as-magic is done in so many places (Ventus is another favourite), and I usually enjoy it, but I felt that in Elder Race it was pulled off in an unusually elegant way.
As you say 'social media' is not a good category, we should specify exactly the things that are concerning. Here are the ones where I'm concerned about their effect on young people:
1. a user is shown new content based on extensive profiling and a secret algorithm that the user does not control
2. a users activity can be discovered and tracked by people that intend to take advantage of the user
3. the operation of the site is optimised for addiction (or more euphemistically "attention")
I absolutely don't think that a book club or a kids own website comments or person to person chat systems should be included in the rules.
Note - I'm not saying these things should be banned, just that I think it's reasonable to restrict their use to adults.
…why do all of those things happen? to sell paid digital advertisement. remove that incentive and I suspect the “social media” problems largely go away
In reality, a large enough group of people on the internet starts to turn sour. Especially with anonymity. Especially without a specific purpose like a book club. Especially without moderation.
Small groups where you know everyone is where it’s at. To avoid internet stalkers and bullies, and for general quality of the community。
Our brains are built for small communities, not billions.
I suspect the reasoning was similar to the reason Tesla bought Solar City or X.ai acquired the site previously known as twitter. Pure unvarnished investor value.
The premise of the steps you've listed is flawed in two ways.
This is more what agentic-assisted dev looks like:
1. Get a feature request / bug
2. Enrich the request / bug description with additional details
3. Send AI agents to handle request
4a. In some situations, manually QA results, possibly return to 2.
4b. Otherwise, agents will babysit the code through merge.
The second is that the above steps are performed in parallel across X worktrees. So, the stats are based on the above steps proceeding a handful of times per hour--in some cases completely unassisted.
---
With enough automation, the engineer is only dealing with steps 2 and 4a. You get notified when you are needed, so your attention can focus on finding the next todo or enriching a current todo as per step 2.
---
Babysitting the code through merge means it handles review comments and CI failures automatically.
---
I find communication / consensus with stakeholders, and retooling take the most time.
One can think of a lot of obvious improvements to a MVP product that don't requre much regarding "get a feature request/bug - understand the problem - think on a solution".
You know the features you'd like to have in advance, or changes you want to make you can see as you build it.
And a lot of the "deliver the solution - test - submit to code review, including sufficient explanation" can be handled by AI.
I'd love to see Claude Code remove more lines than it added TBH.
There's a ton of cruft in code that humans are less inclined to remove because it just works, but imagine having LLM doing the clean up work instead of the generation work.
I think the cycle is because people forget how destructive war is for all sides, how much human wealth is thrown away in order to achieve enormous human misery. If it's happened in recent memory, people are reluctant to let those who think they might benefit from it to pursue it. The more time that passes, the easier it is to distract people from the misery and the easier it is to persuade people that it's justified.
Phones are never going to run the largest models locally because they just don't have the size, but we're seeing improvements in capability at small sizes over time that mean that you can run a model on your phone now that would have required hundreds of billions of parameters less than 6 years ago.
reply