One of the big things that enabled extended context to work was training techniques for extending LLMs with reasoning, part of the thing he was saying wouldn't work.
> Maybe the new tattoos are just like being racist or something, but that’s hard to do when your heart isn’t in it and they will eventually find some way to absorb that.
> What exactly was he dead wrong about that is proven by any of this?
He said as you need more and more tokens models will fall apart because each additional token is a chance for a mistake and they will just exponentially fall apart. But in practice models have learned to identify and self-correct mistakes and if you look at the graphs more inference reasoning tokens almost always give far better accuracy.
The scaling with reasoning models is more and more with things like verifiable rewards (coding and math), in line with bitter lesson and also Sutton invented lots of modern RL.
By bunch up in a ball I just mean assume the fetal position when powered off, not wheels. They wouldn't have to take up more space in house than a large piece of travel luggage on a shelf.
Probably related to things like "The Nazis utilized data from routine censuses, tax returns, and municipal police registrations. In Germany, and in occupied countries like the Netherlands, this information was systematically organized. In some instances, IBM technology (via Dehomag punch card machines) was used to tabulate and sort census data to identify individuals of Jewish descent."
It doesn't need to have a sleeping surface or stay sprawled out when it isn't in use. It could bunch up into a little ball and fit in the corner of your ceiling. And it doesn't need to be the size of an adult to do most household stuff, some of the unitree ones are really short in stature, a foldable step stool for reaching upper cabinets or changing lightbulbs is probably enough.
A bigger issue is whether it can really be as safe, not trip over wires, throw the baby in the trashcan, start a fire trying to make a cup of coffee and that kind of thing. Beyond accidents, lots of companies are talking about hooking these up to LLMs for planning that have horror movies in their training sets.
In some fields like comp sci, when code isn't given but the paper describes the approach, LLMs do help with the reproducibility crisis: you can ask it to reproduce the result through reimplementation by reading the paper.
If it fails you may have to double check it did properly reimplement it, but if it succeeds you do get a reproduction.
Why have they talked about this for a long time? They predicted date of code maxing out, and did so not from fitting a sigmoid or something but they predicted it would max out right during a steep part of the slope?
To add to this: one of Anthropic's big quality problems this year was claude code began leaving out the reasoning when a chat was idle for X period of time and was revived. They said it was to avoid delay and make it feel better but it was during a huge capacity crunch so probably just something they felt compelled to do. Without the reasoning in the history model performance degraded greatly when the chat continued.
For one, 1926 we didn't have just "commoners" and "aristocrats." Two, people would save for special occasions at places like fancy restaurants, even if not to the extent of today, and even if more constrained to larger cities.
> The magic of automation, and in particular the
magic of an automatization in which the devices
learn, may be expected to be similarly literal-
minded. If you are playing a game according to
certain rules and set the playing-machine to play
for victory, you will get victory if you get any-
thing at all, and the machine will not pay the
slightest attention to any consideration except
victory according to the rules. If you are playing
a war game with a certain conventional inter-
pretation of victory, victory will be the goal at any
cost, even that of the extermination of your own
side, unless this condition of survival is explicitly
contained in the definition of victory according
to which you program the machine.
> ...
> In short, when there is
a war game to program such a campaign, there
will be many to forget its consequences, to ask
for the £200 and to forget to mention that the
son should survive.
> While it is always possible to ask for something
other than we really want, this possibility is most
serious when the process by which we are to
obtain our wish is indirect, and the degree to
which we have obtained our wish is not clear until
the very end. Usually we realize our wishes, inso-
far as we do actually realize them, by a feedback
process, in which we compare the degree of
attainment of intermediate goals with our antic-
ipation of them. In this process, the feedback goes
through us, and we can turn back before it is
too late. If the feedback is built into a machine
that cannot be inspected until the final goal is
attained, the possibilities for catastrophe are
greatly increased