I've had similar issues for my site AppGoblin. I offer 1000 daily free API calls and have had rashes of various bot like activity signing up to use the free API calls. In the end I had to add CloudFlare inbetween the signup process, which is a huge pain as I know that blocks many normal people on VPNs or in smaller countries.

The bot users I found were never particularly sophisticated, using certain name or email elements that were similar and semi obvious. Still, they came from all different domains, many with realish looking vibe coded sites that only looked faked once you really poked around (no contact, or no linkedin, extremely vague, no more than one or two pages etc). And of course the most obvious thing, that one or two I wouldnt' think much of, but seeing 12 similar name/site/email combos in a day and you realize it's coordinated.

ah yeah, i've also tried building some ai services and offering a free tier to the first x users just to have bots filling them up... its such a shame. Dont like the whole cloudfare or human verification ux of a site but maybe those are just the times we're living in

"The legislation would permanently prohibit artificial superintelligence, which it defines as a system that exceeds human cognitive ability or has sufficient capabilities to plan and execute the destruction or disempowerment of humanity."

Wow huge miss by Bernie. I mean in many areas models from a year ago were above and beyond most human's cognivity, or at least in rote memorization of a wide variety of facts.

Not to mention, he says ban, but realistically it means to take away from regular people and the government and corporations with licenses keep using it.

I really do feel like this is going the way of the Second Amendment. I wonder if there is grounds to have a right to use AI as a defensive means.

Maybe Anthropic can pause so that open weight and open source models can fully catch up?

If they really believed their talk, they would open their weights and harnesses so we could fix their obvious insecurities.

Or maybe this is a veiled threat, now that they're messing around with viruses. This whole thing is getting worse by the hour.

if they are such "leaders", they ought to show the way by "leading" without the need for governments to force their hand

I heavily use browser based LLM for bugs I care about. Code that is close to business logic like SQL is all 'by hand'.

Frontend stuff for me, which I'm bad at, I let agents do all of and I test the end result.

Other things are in-between, but time wise it's probably mostly me writing code. Lines of code is the LLM for sure.

This looks great in terms of cost and capabilities, truly pushing the frontier forward in terms of open weight light weight models.

I think some of it comes from that they do not publicly let you see the random seed. So each time you ask the answer is different (like a slot machine) and if they let users use the random seed it would let people much more accurately assess if an underlying model changed somehow (same seed and same input will always have the same output).

Of course the closed Anthropic would never share this, it would definitely take away the 'magic' feeling of the AI

With models there are a bunch of other dials that can be tuned even if the model itself remains exactly the same.

Are those dials set the same across all hardware configurations and clusters? Does model behavior average out the same across different hardware?

There are just too many different buttons that can be set to really trust a provider either not to directly commit fraud, or indirectly commit fraud with system complexity affecting the output.

To my understanding, with batched inference and other "optimizations" you wouldn't get the exact same token predictions even with temp=0.0.

I really hope we get to crest this hardware supply shortages and drop into oversupply.

Across the global volume of 185 million to 220 million laptops per year, where manufacturers were putting in 16GB RAM for the past decade (or longer), we really need 128GB to run local inference for the AI-centric world of the future, so that is an additional 112 GB per laptop - this creates an annual shortfall of 20.72 to 24.64 billion gigabytes (20.72 to 24.64 Exabytes) of DRAM. Just for laptops alone, ignoring the desktop or server or cloud or other markets.

Manufacturing supply is going to take 5 years to catch up. Buy memory stocks (MU/SKHynix) - they will be multiple of their current prices in 5 years.

I spend half my time in Taipei and regularly see young children walking, taking the metro etc. In comparison I recently went camping with inlaws who had no shortage of complaints about the children riding bikes on their rural country road, doing wheelies and what not. To be fair to their point, the posted speed limit is 45MPH and traffic definitely travels well above that. But it's so sad how the US has no room for people not in vehicles.

In Taipei you can see this slowly affecting the city as well. Many new pedestrian "safety" measures inconvenience pedestrians to keep them safe from cars. I'd rather see the cars inconvenienced than the pedestrians, but I wonder if this is just a natural progression.

The one place these seem to not be effective are smaller streets. Small streets seem to keep the cars and pedestrians much more equal.

The vscode chat is great. Not sure if it's also called copilot, but you can plug in any models there and they get sandboxed, tools and link to your code. Great stuff.

I love VSCode built in LLM harness. No extensions, use any model you want, amazing UX for reviews, control of agents. The sandboxing also seems great, I never see models touching anything they shouldn't.

Is this the chat on the side or the agents feature?

I have to admit I'm not clear the difference? It starts with the list of your sessions/agents . Then starting a new "agent" or question or plan. I usually use each agent for a few questions or prompts at most. Occasionally it runs longer than that.

I love using it for the chat question style where I ask it things like:

For these frontend pages with X options collect the database tables for each.

This is of course something easy to do manually, but I like it stepping through a couple functions much faster than me searching around manually and maybe getting lazy and not checking all of them.

Both are great, and analogous to Cursor which has both features too.