nazgul17

Jev's particular architecture is such that context can be filled faster than with LLMs.

But we are not buying oil or gas. We are buying hardware that will generate electricity for decades.

I do hope the US turns around. As I see it, we (western countries) have aligned values and it's better for us all (US included) to stick together.

Could you share more about how you guys set up the rewrite? I am interested in the shape of the software as well as how you achieved (or aimed to achieve) high quality.

Competition is better for everyone

GEN 1.5 demonstrates in-context learning and unseen-before generalisation

Makes sense, but I would hope the study design controlled for this, or designed around it.

My 5 years in school timetabling say: 100% agree!

But that's not what happened here. Here, you were trying to bring the pepper over, got inspected and somehow got rid of it because you were able to be found out.

Just a few years back, wind turbine software was hacked. This is serious.

Serious governments should (and hopefully will) scramble to use AI to discover and patch as many software vulnerabilities in infrastructure. They can do it with the agents red teaming and using ultimatums to companies to fix each vulnerability they find. Not doing so is equivalent to exposing your flank to disruptions in peace time and to attacks in war time.

There's alignment (trained in the weights) and there are constraints (in the "server harness"). My take is that this model was not yet aligned and had no constraints

From what I understand, whether you will struggle depends a lot on whether you eat well, sleep well, drink, smoke, and exercise.

My dad played futsal until the end (~70 yo, car accident, be was perfectly healthy otherwise); I met one of his classmates from school, he looked 20 years older, no exaggeration. My dad looked like he was barely 50.

Note: how old someone looks is very biased by their "generation", as people tend to drink and smoke less these days.

That was DeepSeek OCR, not a DeepSeek lineage LLM. If the idea is introduced in LLMs, then you're right. But Gemini is not doing that, not yet. This is something I literally discussed with Claude last week, but took its word for it.

I think the more you know of how (many) things work, the slightly better you'll be at using them. From dishwashers to CPUs, from car engines to watercolours, from guitars to kitchen knives... You get the gist. Once you internalize a model of the thing, it becomes closer to an extension of you than a tool. You drive it better and with less friction.

To play the devil's advocate, though, you wouldn't drink espresso at 80 degrees, no? Provided the cup is warm to begin with, you could brew at 50 degrees and still have perfect temperature espresso for drinking? Those 30 degrees are bound to save a bunch of energy, I'd think.

Not GP, but I think the point was that no extensions => no ad blockers => major malware vehicle unlockable, short of disabling JS

Isn't this a consequence of Linear No-threshold, a model that most policy is based on, that says that the bad health consequences of radioactivity are linear wrt the amount of radiation, and that there is no threshold whatsoever under which the radiation is harmless?

A model that is not based on science, given we know that cells have repair mechanisms? Jesus, even bananas are somewhat radioactive, so why are they being sold if any radiation is bad?

Thankfully, it seems the winds are changing in the US, where LNT is being replaced by science based models by regulatory bodies. I hope the rest of the world swiftly follows. The amount of deaths and damage and suffering and money that could have been avoided is mind boggling. If I imagine an alternate history where starting 70 years ago (even just some of) the money invested on fossil fuels or used to subsidize them had been directed to nuclear, and what the state of science today could be, what the state of the air could be, the number of floods, tornadoes, lung cancers that could have been avoided, forced displacements that could have been avoided and subsequent depressions and suicides (see Fukushima), my blood boils. It truly is a mistake of disproportionate scale, and a matching shame.

The US is a single country. Russia is not on the US' doorstep. The US has its own oil. The US prints the world reserve currency.

Apparently (heard from several sources, haven't tried myself), sticking with it for a few weeks causes your gut biome to adapt, then you gas anymore.

Can you elaborate?

In a scenario like this, where we can only know if the code has bugs, not if it doesn't, isn't survival analysis a more appropriate statistical technique? I.e. a technique where time is a first-class citizen.

By the way, I did find this a bit hard to read but, as instructed by OP, I'll go fuck myself.

For what it's worth, I find AI written prose easy to read, and am annoyed by all the constant HN comments which just point out the author was AI, without anything else substantive to add.

Just thinking aloud.

Could it be that the "stories" we consume (books, movies, TV, games) fill our heads (as in, focus our attention) with ideas that don't overlap much with family life? It's all about single people doing cool or fun stuff, with the occasional tired parents.

I wonder how hard it would be to give credibility to the hypothesis (yeah, "verifying" it would be much harder). What's the correlation between "percentage of time spent on non family focused stories" and "children per woman born in the following 5 years"? Sounds like a super noisy signal. Maybe averaging by country or even city could strengthen the data.

Exponential decay isn't impossible to manage if the exponent is reasonable (i.e. closer to 1, i.e. if the average number of children per woman was closer to 2.1). It's just going down too fast.

The announcement of GPT 3, hands down. That's the day that my mind was blown.

Everything after that has been (genuinely significant) incremental improvements. But that announcement was a qualitative step up: we got ""real"" AI that day, something that could pass a Turing test (as common sense envisioned it, without all the caveats added once we learnt of the genuine limitations of LLMs).

I don't see these QAT models on Edge Gallery; just the BF16 models are there. Is there anything I am missing?

A recent episode of the MindScape podcast seems very relevant here:

Andrew Guthrie Ferguson on How Your Data Will Be Used Against You https://open.spotify.com/episode/4FOwAWkB0Bu00EpxmE97qB?si=V...

It seems that the cutoff date for all models is stuck at some point before AI generated content started being pervasive.

It's just a metric. If it can find a needle in a 1Mtok haystack, then it's likely good at coding within a 200Ktok context (or whatever, insert your number here, I'm just trying to make a point)

Is my understanding right that the system is limited by the capability of teachers to solve their own problems? The TrueSkill of teachers doesn't seem to increase all that much, but I suppose TrueSkill works like that if the whole population gets better.

You can have both of you provide an error code that your support team can look up in the logs to find the complete error message.

Ahem did you know David wasn't always 100 years old, and that in the past he worked on the field?