HN user
Who else? The people actually holding the hostages, perhaps.
Because you surely mistakenly implied here that All Gazans categorically are responsible for the hostage-taking and "deserve" to be killed as a consequence of that.
Your use of language betrays your feelings. You are disgusting.
They definitely were a thing all the way back to the first browsers.
I am not disputing that ads were a thing. I am not disputing that ads were common.
I said that there were a lot of sites that chose not to run them.
They all wish they had the viewership for ads.
This is just not true. Like, c'mon man, the very site you're on right now takes this approach.
So, obviously, ads were the norm back in the day. The author had to be wearing several rose tinted glasses when writing that.
But the author isn't entirely wrong. There were/are a lot of websites that simply did not run ads. Hosted not for money, but "for love of the game".
This is something that was lost with the shift to exclusively platform-based hosting. A facebook page or subreddit simply is never going to be ad-free in the way that a lot of former or legacy forums were and are.
Being older makes a person more likely to be homeowner, they got the causation backwards.
No.
Being a homeowner doesn't grant one political influence. Being old grants one political influence.
It's the correlation of age and homeownership that means homeowners have the political influence the push through policy that drives up real estate prices.
Non-homeowners have political incentives all the same. If only just to oppose those very homeowners' policies. What they lack is the political influence to make it happen.
The problem is that it is essentially impossible for a journalist to exist in the western world and not have heard of the criticism about how cops' actions get reported.
The term 'past exonerative tense' is dated to 1991.'"Mistakes were made" was popularized by Nixon.
To continue pulling this nonsense is wilful ignorance on the journalists' part, and effectively equivalent to bad faith.
You don't need a new model. The trick of the technique is that you only change how tokens are sampled; Zero out the probability of every token that would be illegal under the grammar or other constraints.
All you need for that is an inference API that gives you the full output vector, which is trivial for any model you run on your own hardware.
No one wants to admit this point so we keep arguing about whether we want to leave addicts to die in the street or in a crowded crack den. Neither really solves the problem.
That is correct, yet at the same time: Society as a whole refuses to give these people even the kindness of a roof over their head.
They need better care, yes. But if people won't even agree that these people shouldn't freeze to death in winter (or overheat in summer), talk of funding better care is off the table.
Christ, Fox News had one of their guys outright suggest they be euthanized. The bar for discourse on homelessness is in hell right now.
Somehow the homeowners almost always win against the renters in this political tug-of-war.
Demographics. Homeowners skew old, which gives them a bunch of advantages in enacting their political power. Higher turnout, baby boom giving them numerical superiority, and the time advantage of being able to enact policy decades ago.
In the US, this is supplemented by matters of race, where because of past redlining policies, "pro-homeowner" policy (esp. suburban single-family-homes) in the last half-century has been a way to primarily benefit white people.
There are some implementation concerns, but the real answer is that it is an ideological choice.
The AI companies believe that these kinds of grammar mistakes will be solved by improving the models. To build out tools for grammar constrained inference like this is to suggest, on some level, that GPT-N+1 won't magically solve the problem.
The deeper level is that it's not just simple grammar constraints. Constraining to JSON is a nice party trick, but it opens the door to further ideas. How about constraining to a programming language's grammar? Those are well defined, you just swap the JSON grammar file for the Java grammar file, job done.
We can go further: Why not use a language server to constrain not only the grammar but also the content? What variables and functions are in-scope is known, constraining a variable reference or function call to one of their names can be done with the same techique as grammar constraints. ("monitor-guided decoding", figured out back in 2023)
Entire classes of hallucination problems can be eliminated this way. The marketing writes itself; "Our AI is literally incapable of making the errors humans make!"
What many AI developers, firms, and especially their leaders find grating about this is the implication. That AI is fallible and has to be constrained.
Another such inconvenience is that while these techniques improve grammar they highlight semantic problems. The code is correct & compiles, it just does the wrong thing.
Implementation preference.
is masking expensive?
It's not expensive per-se; A single element-wise multiplication of the output vector.
The real "expense" is that you need to prepare masks for every element of your grammar as they are expensive to recompute as needed; LLM tokens do not cleanly map onto elements of your grammar. (Consider JSON: LLM tokens often combine various special characters such as curly braces, colons, and quotes.)
This isn't that hard to compute, it's just more work to implement.
AI criticism and pushback.
When you say "AI cannot do my job, [insert whatever reason you find compelling]" Execs only hear "I am trying to protect my job from automation".
The executives have convinced themselves that the AI productivity benefits are real, and generally refuse to listen to any argument to the contrary. Especially from their own employees.
This impedes their ability to evaluate productivity data; If a worker fails to show productivity, it can't be that AI is bad, because that'd mean the executives are wrong about something. It must be that the employee is sabotaging our AI efforts.