You will still get hallucinations. With RAG you use the vectors to aid in finding things that are relevant, and then you typically also have the raw text data stored as well. This allows you to theoretically have LLM outputs grounded in the truth of the documents. Depending on implementation, you can also make the LLM cite the sources (filename, chunk, etc).
The approach that has worked for us in production is correction during generation, not after.
The model verifies its output against the rules in the prompt as it generates and corrects itself within the same API call — no retries, no external validator. If there are still failures the model cannot fix at runtime, those are explicitly flagged instead of silently producing wrong output.
This does not mean hallucinations are completely solved. It turns them into a measurable engineering problem. You know your error rate, you know which outputs failed, and you can drive that rate down over time with better rules. The system can also self-learn and self-improve over time to deliver better accuracy.
I think generally, SFT is like giving the LLM increased intuition in specific areas. If you combine this with RAG, it should improve the performance or accuracy. Sort of like being a lawyer and knowing something is against the law by intuition, but needing the library to cite a specific case or statute as to why.
This patent came from the design department. Some dude looked at the logo and said, “Huh, we could make a screw from this.” The legal team then took it and made a design patent.
If you guys want, I can message him and ask about the grand conspiracy behind it but you might be disappointed.
Just wanted to point out that California, for being barren of manufacturing and hostile to it, is actually the largest manufacturer in the USA in terms of people employed and economic output.
It's also multiple times the size of most states. Per capita is probably a better metric, where it ranks 34th.
> and economic output
I can't find the raw data, so I'm not sure what they're counting. It says "computer and electronics" is highest, which is 2x above chemical, which makes me very suspicious. I work in manufacturing that happens 100% in China. Would I be counted? Are they counting AMD or Qualcomm, where their chips are manufactured in Taiwan (or Arizona)? How much of this is military/government contracts (which aren't subject to many of these pressures)?
That's because it's so much more populous. That link shows that CA is 34th in % of workers employed in manufacturing (similar to TX and MA). CA is tops in many, many absolute measures because it has almost 10M more people than the next state, TX.
Just that "largest manufacturing state in the USA" isn't a very helpful measure, since the largest state in basically every measure of development. This tweet shows the same problem with absolute numbers giving misleading appearances: https://x.com/xkcd/status/1339348000750104576:
There are more Trump voters in California than Texas, more Biden voters in Texas than NY, more Trump voters in NY than Ohio, more Biden voters in Ohio than Massachusetts, more Trump voters in Massachusetts than Mississippi, and more Biden voters in Mississippi than Vermont.
There's network effects in manufacturing and not so much in presidential voting, size actually does matter. It's one of the things people always rave about with China.
I'm somewhat surprised that the other response to me thinks the market that is ~1/10 the size of the other is obviously a better place to work. I guess cost of living can have a big impact.
Whenever I hear VRML, I am reminded of this Aptiva commercial from the 90's with the "3D San Francisco" site. Was always bummed as a kid that I never actually could find it online. https://www.youtube.com/watch?v=ROOSUiLEesA
It’s interesting to see how optimistic and forward thinking architecture for education could be during this time period.
It seems by the 1990s, many attributes that were seen as desirable (open air, windows, single level) were at some point abandoned in general, leading to windowless boxes and long wheelchair ramps between classrooms (at least this was my experience in Los Angeles).
