Note that the model is based on RoBERTa and has only 125m parameter. It is not competing against any of the new popular models, not even small ones like Phi or GeMMa.

It’s also not meant to be a generative model - only to be used as an encoder model (they list retrieval as a potential use case )

Given the current state of LLMs, I am not even sure this qualify to be called an LLM.

second opinion - BERT family are transformer-based, and that is a big threshold right there.. secondly I am not sure that two one-minute comments could capture what exactly went on with fine tuning or graph-based methods of constraint or whatnot.. with respect to the fitness of the production tools for intended purposes.

Pedantic remark: HTML 4.0 is an SGML profile, but HTML 5 isn’t anymore.

They have fixed the image now.

Thanks. Yep it looks correct now. :)

For German there is an international body. Regional language differences are officially acknowledged. For example, Swiss High German has no Esszet and in Austria January is spelled "Jenner" instead of "Januar".

Eszett, Jänner

It doesn't work. There is no encoding for "OO".

You are right! Sorry totally overlooked this :) Thanks.

Wow good eye?

> Microsoft keeps paying the bills

For those who may not know it: SPJ was employed at Microsoft Research for a long time. He recently moved to Epic Games, where he co-developed the Verse language.

Probably because some people find it funny.

Because there is no _intent_ to slander.

So if I send an email to your boss saying "occamrazor is a <blank> molester", I'm fine as long as I say that I didn't mean to cause any harm?

You are describing the actual malice standard which only applies to public figures. Also, an accusation of that type would be considered "defamation per se" which would mean that there is not a need to prove damages. According to my layman understanding.

https://en.wikipedia.org/wiki/Actual_malice

https://www.findlaw.com/injury/torts-and-personal-injuries/w...

HN needs a button to "ask a professional" to resolve a question in a thread, be it a doctor, lawyer, accountant, physicist, or electrical engineer. ;]

Ah, that's a good point, thank you!

am I also fine if I negligently write and keep using an algorithm to detect and report people for being molesters when I know full well that it's actually shitty and I regularly receive reports that it's resulting in false positives?

That's facebook's situation. Even though they've got hundreds of billions and could easily afford to, they don't want to pay humans to verify these things and they know that innocent people are being hurt by their terrible lazy code, they just don't care.

I suppose you'd have to prove that they knew the algorithm was bad. They'll just say "we're trying our best, but it's hard". And the truth is that it probably is hard. The question is whether knowing that your algorithm is imperfect is equivalent to intent to harm.

Sure there is, they're a bigcorp and did what they did. Don't let companies get away with crimes you and I wouldn't just because they're too disorganized.

Depends on the legal jurisdiction. No intent required in England for example, just proof of serious harm

Hijacking the right click is useful for web apps like word processors, text editors, spreadsheets, image editors.

Because total dollar cost is dominated by major hurricanes, which are rare, therefore any underlying trend would be invisible under the stochastic noise.

Frequency at a moderate severity level is much more stable and can give better insight.

As other posters already pointed out, the severity of the events should be adjusted for inflation, building characteristics, and total exposure.