Professionally, I'm forced to use CoPilot, which is so eager to please me that it's useless, and Claude which is fairly good at writing code but hallucinates data when I ask it to organize my notes.
Personally, I'm using Qwen, Phi, etc. to squeeze as much performance out of constrained hardware as possible. They're pretty good at searching my personal files which you mentioned in a previous comment.
> the best use, arguably, of an LLM is as a very fast search index for common information
I'd counter forming a rebuttal against a search result would be fairly trivial. I.e an AI summary for "are oranges poisonous" would say "Moms Against Big Citrus says oranges are poisonous, but the Association of Orange Groves says they are not." When used for search, LLMs are a source for sources instead of the source of truth itself. The user decides which source to believe instead of deciding to believe the LLM itself.
> you might be fortunate to be in a position where nobody is relying on your velocity to ship products at competitive speed.
Admittedly the biggest drawback. AI is pretty terrible at my profession (electrical engineering, regardless of recent demos that are horrendous when examined under a microscope), but the software engineers I know haven't hand written industry code in months.
> If its the difference between the information being available at all
That's the point. The solution should avoid information not being available. Requiring login will incentivize bots to create spam accounts and move the battle to a new frontier, hurting real people in the process.
I can walk into a library, pull a book from a shelf, sit down, and read it front to back. No library card needed. Only need one to take a book home. Completely different scenario.
Only because there's no way for bots to exploit the system since it's in the physical world.
If thousands of robots suddenly showed up at your local library and started hogging all the books so nobody else could use the library, you can be sure a library card would be required to even enter.
You clearly do not live in an area with a large homeless population. Many libraries are effectively under-resourced homeless shelters. Yet, no requirement to have a library card to get in. Information is still open to the public.
1. I have never been to a library that has banned any class of people. Let alone specifically homeless people.
2. I don't know the laws of your jurisdiction, but banning a whole group of people from a public place would, by definition, not make it a public place.
3. The overwhelming majority of librarians view libraries as a service for everyone, including the homeless[0].
Software moves so fast that its actually an anti feature. Libraries and SDKs measure lifetime support in years, but industrial hardware support *at minimum* is a decade. A lot of these new startups dont understand that established industry wont use them unless they known this shiny new system wont ve abandoned in a few years.
On the other hand, softwate devs trying to quickly make a product have something to get them going.
A lot of datasheets are inaccessible without signing an NDA. Even if you're a VC-backed enterprise, some unnamed chip vendors will tell you to pound sand if you're less than 8 figures of annual revenue.
Worse, many recommendations in datasheets were written in the 80s, overkill for many applications, or completely wrong and won't be updated until an errata is published 6 months later. Given current workflows, I don't see how AI would help designing a PCB with an alpha chip that's doing anything remotely novel.
More to the articles point, AI designing ASICs or other Verilog/HDL defined components could be very interesting use case
Not necessarily true, most recently I was working with an USB 3.1 chip that anyone can buy off DigiKey, but I had to call contacts at 4 seperate companies just to get firmware to make superspeed work. We werr just too small of a fish to make the chip vender care.
This is becoming more true as a larger percentage of hardware infrastructure is occupied by the datacenter economy.
I'm trying to find your capabilities. I.e trace and space minimums, hole to edge minimums, min annular ring, etc. Are you not at that stage yet? Will you ever publish a standard spec?
Personally, I'm using Qwen, Phi, etc. to squeeze as much performance out of constrained hardware as possible. They're pretty good at searching my personal files which you mentioned in a previous comment.
edit: typos
reply