As someone who loves programming, I think the distinction is overstated. Part of the reason why doing what I love is slow, is because I instinctively (try to) verify as I go.
I think it's kind of happened already. All the time we see news of politicians or famous people having their very old photos, comments, or reddit accounts found with distasteful takes. And it seems they can mostly just handwave it away with "Hey that was 10 years ago and I wouldn't make those comments today" and nothing seems to come of it.
As an AI engineer I made mysefl a promise when started a company:
"You will only use Ai, if it really useful for your customer" IT wasn't from 2021 to 2025 but then LLM reached a point that now the experience for my customer is way better using a completely different product that is a conversation instead of a classic Airbnb search. What do you do when you discover that? You change you product
Also we do not use AI "buzzwords" for the marketing at all, if you go to our website https://www.teamout.com/ we do not even mention it, we focus on the experience you will get at the event.
I will repeat here again the same comment I made when they posted their constitution:
The largest predictor of behavior within a company and of that companies products in the long run is funding sources and income streams, which is conveniently left out in their "constitution". Mostly a waste of effort on their part.
"the code" is the issue. This isn't building code like protecting earthquakes, electrical, fire safety. When people hear "the code" -- they think critical safety measures.
Activists came up with a doorjamb law to require bike lanes, ADA and road diets on every block that receives repair. That's just not practical in LA. You can't punish drivers into riding a bike to work.
this is coming from someone who biked and bussed to work in LA for 10+ years.
It's reckless policy that will only undermine your efforts and never reach the desired outcome.
The way this article was written, is the standard way these kind of US pop science articles have always been written. It's LLM that absorbed that, not the opposite.
A number of cities in Northern California are doing just this. We have at least 12 high density projects being built in Santa Cruz and we are a small city.
> Everything you wrote above was said about PDAs in the 90s
No it wasn't. PDAs were seen as crappy little computers, but the applications were obvious because the bigger much more impressive computers were everywhere by then. There was no question about the value of personal computing anymore.
Everything regarding probabilistic AI is either about optimization or trading off costs. All such applications are intrinsically and perpetually lost in the weeds. The use cases aren't new because "generative" is a marketing buzzword desperately trying to cover up what is actually just "imitation".
AI makes what was already possible more accessible. It is useful, but not a revolution for the layperson or even most businesses. People are in awe of the money being exchanged, but are also in denial that it's almost entirely defense spending.
We’ve been building in this space for a while, and the issues listed here are exactly the hard parts: session connectivity, reconnection logic, multi-session UX, and keeping state coherent across devices. Especially when it comes to long running tasks and the edge cases that show up in real use.
I’ve been exploring why LLMs "break" during inference. Most current hallucination detection methods look at the final text (semantic analysis) or use another LLM to double-check (self-consistency). These are effective but extremely slow and expensive.
SIB-ENGINE is my attempt to solve this at the geometric layer. By monitoring the "Anchor Drift" (how hidden states deviate from the prompt’s latent trajectory), I found that hallucinations often manifest as a structural instability before the token is even sampled.
The Numbers:
Recall: 53.89% (It catches about half, but it's consistent)
Precision: 88.52% (Low false-alarm rate is my priority)
Overhead: <1% (Running on an RTX 3050 with 4GB VRAM)
AUC: 0.8995
I've released a Lite version (1-axis) on GitHub so you can see the fundamental logic and run it on your own machine. I’ve also included the raw_logs.csv from my N=1000 test run on Gemma-2B for full transparency.
I’m particularly curious if anyone here has experimented with similar geometric approaches or has thoughts on how this might scale to 70B+ models where the latent space is significantly denser.
I think CA has bigger issues to tackle than repealing Prop 65 (IIRC, that's the one about carcinogen warnings). The other, may be happening in some places, but having lived here 20+ years in the Bay, I haven't seen even one example of either.
Ironic because IEEE Spectrum has an anti-LLM policy. So your complex about LLM writing styles has indirectly caused you to stop supporting genuine prose.
Seriously there's no LLM stuff in here. Only emdashed which were used in journalism decades before AI was even a thing.
These days you can rewrite everything yourself for very cheap. So this is `mcporter` rewritten. I prefer to use Rust personally for rewrites. Opus 4.6 can churn it out pretty quickly if that's what you want.
And effectively it did: many people are kept in their place by the combined pressure points of debt and employment to stay (barely) afloat.
This is of course nothing compared to the cruelty of real slavery but the effect is much the same, a lot of people are working their asses of for an upper class that can ruin their lives at the drop of a hat. That there are no whips involved is nice but it also clearly delineated who was the exploiter and who were the exploited. That's a bit harder to see today.
As I understood the grandparent post, the idea is that a highest-level university should 10× its student throughput, and 9 other, lower-level universities would be made redundant by that.
This would make sense of all an elite university did were providing elite-level education. Of course exclusive schools provide other benefits, often more valuable for the target audience than the education proper: a highly filtered student body, networking and bonding with the right, upwardly mobile people (either mega-talented, or just smart kids of rich and influential parents), a luxury-grade diploma that few can afford. Maybe you could theoretically 10× Stanford or MIT, but likely not Yale.
Love this, will try without any AI assistance and see how it goes. Thanks for the feedback, I'm starting to write after all, who knows, maybe I'm an even bigger slop machine myself...
I've added your post, as well as the 2018 paper on Behavioral Cloning from Obervation (BCO) to my reading list.[a]
So far, I've only skimmed the 2018 paper, but it's already evident that it's well-written. BTW, BCO is a really good, easy-to-remember name and acronym.
Thanks! A graph visualizer is interesting, but it’s not really the goal here. There are already good tools that map entire schemas. DBTree is more about focused, node-by-node data exploration, without loading or visualizing everything upfront.
It relies on actual foreign key constraints. If there’s no FK defined, it won’t infer the relationship. The chevron is shown based on those constraints, and the related rows are only queried when you expand the node, which keeps performance under control.
Based on this and your other comments, including the one that’s no longer visible: Please phone a friend. Or find a professional to talk to. I say that with nothing but compassion.
