When we say "safe" it's a regulatory statement about _certainty_ not about any given person's activity. We know pasteurized milk is safe because the process produces a high probability of a safe product.
When we don't do that, it's called raw. From there, we don't need to investigate anything else, whether it's 1 in 100, 1 in 10, or whatever. We know that because it's unprocessed, it's unsafe.
It's always curious when people bring anecdotes to a discussion like this as if what their family did with raw milk is perfectly emulated everywhere.
Same thing happened with surgery in the early century: doctors wouldn't wash their hands because they had some base assumptions about what caused diesease.
In the end, countering these anecdotes rarely work.
I think the problem is a stochastic one: More options seem to exist for this technolgy to abuse humanity via it's "owners" than do it to democratize anything. It's not like it's helping to wage war, molify the public and entraining pre-existing racist for the last decade.
These are all things happening today via AI, so really, this is an argument thats like, entropy. There's always way more ways in which things fall apart than they build to stability.
Being optimistic seems more like religiousity than any real accounting of the current system you're operating in (unless you're a billionaire).
So on the one hand, this is an absurd ban; on the other hand, whenever this corrupt USA government does anything this absurd, it usually signals the start of the kleptocrat activities.
So, whose going to come rescuse us with the CLEARLY superior technology that ABSOLUTELY DOES NOT implement the very thing the FCC is trying to protect the SMALL american from?
Practically all routers are made overseas. So unless new factories get built ASAP the U.S. is all but guaranteed to have a router shortage and a thriving grey/black market
That's missing the aspect where exceptions to the ban can be granted by the DoD or DHS, so in practice the outcome will be that effectively all routers need to appease the national security apparatus before getting FCC approval.
The United States has many close allies who manufacture routers. Seeing as how we already share intelligence and military technology, banning their routers seems... inconsistent.
The part that will make it absurd is going to come when Trump suddenly greenlights some made-in-China routers because the CEO responsible made a "donation" to a "charity." Probably the presidential library.
reads like chatgpt talking to claude about imaginary things.
While this may be a real human reality, the way it's presented is in the golly-gee-whiz, I'm just a farm-folk engineer.
If you meant this to be convincing, it's not. It looks like copy-paste-find-replace of all these other tech blogs where they found $SHINYNEWEVIDENCE of $MODUS_OPERANDI and you should too.
I don't understand what your comment is about. You don't like his style, you disagree with the evidence, or the conclusion?
The author is Brad Feld [1], who wrote checks to thousands of startups, wrote a dozen books, and advises a bunch of founders. He's talking about his personal experience observing the shift in the typical profile of a startup entrepreneur.
I think his perspective is very valid. For the past 20 years we assumed (and confirmed through empirical evidence) that having a technical co-founder was critical for the success of a startup.
This era is getting to an end, and the next 20 will be radically different in the next 20. You'll probably still need human engineering skills to scale, but getting from 0 to 1 will depend much more on taste than how good you are in <language X>.
Yeah why are all these people considering evidence in the first place? Have they no faith? $ONLINEGURUINFLUENCER said AI can only produce slop so it must be true!
What does Linux kernel Devs know about real software development anyway? [1]
I think it's basically equal to End of Line when it comes to an LLM. It means they have nothing else to add, there's zero context for them to draw from, and they've exhausted the probability chain you've been following; but they're creating to generate 'next token' and positive renforcement is _how they are trained_ in many cases so the token of choice would naturally be how they're trained, since it's a probability engine but it doesn't know the difference between the instruction and the output.
So, "great idea" is coming from the renforcement learning instruction rather than the answer portion of the generation.
So, there's things you're fighting against when trying to constrain the behavior of the llm.
First, those beginning instructions are being quickly ignored as the longer context changes the probabilities. After every round, it get pushed into whatever context you drive towards. The fix is chopping out that context and providing it before each new round. something like `<rules><question><answer>` -> `<question><answer><rules><question>`.
This would always preface your question with your prefered rules and remove those rules from the end of the context.
The reason why this isn't done is because it poisons the KV cache, and doing that causes the cloud companies to spin up more inference.
reply