I think people could try social experiments and compare the productivity differences resulting from different management styles on LLMs instead of humans.
If it worked 99% of the time, compliance would be nil, but you don't want these drivers to be affected 1% of their commute
Even if a person would be compliant, if they wore this most of the time, they wouldn't be accustomed to the varifocal lenses while driving, guaranteeing that whatever other solution than these autofocusing lenses they select for driving, they won't be adapted to them whenever they drive.
Uploading human minds will happen faster than medically maintained immortality.
For the latter you need to solve against any kind of failure for all organs.
For the first you only need to be able to recover the important subset of brain state.
People don't appreciate the amount of traces events, things or creatures leave behind. From footprints, to DNA, to ...
Most of the brain state is actually for maintaining the brain tissue cells, and general cellular homeostasis.
A frequently used synapse will have multiple characteristic differences with a less frequently used synapse. To transport this specific facet, there will be multiple markers, but we only need to be able to record one of them, then the rest become redundant. The same can be done for other facets: signal speed and delay for a specific axon/dendrite for example, they will all have multiple correlated measurable quantities, from RNA molecules, to protein concentrations.
You only need one of many indicators to collect that facet of brain state.
Knowledge transfer will be a testable hypothesis: how do putatively uploaded mice respond to mazes they know from meatspace life exposure (walk without detours to prize) versus mazes they don't (walk exploratorily before finding prize).
Once this becomes possible higher ranking prisoners of war may be uploaded for extracting the episodic memory, training locations, training contents, organizational structure, evidence of foreign support in equipment or manpower, observation of war crimes, and if ICC doesn't accept such evidence, PoW observation of PoW-friendly soldiers commiting war crimes, to blackmail these soldiers, potentially during active combat, turning active enemy soldiers behind enemy lines into your agents, many war crime observations will be war-criminal-instructions from superiors, so now you can blackmail them...
The list goes on and on...
Also domestically: conspiracies against a fraction of death row inmates, some will consent to uploading their brain if it means potential continued existence when deemed innocent after investigating episodic memory. If the inmate is aware of some identities throwing them under the bus, those can immediately be questioned, with those discovered not cooperating being uploaded against their will. Quickly undiscovered conspiracy rings will lobby against uploading for discovery, or killing those who know to much, potentially incriminating themselves further.
Imagine the elimination of unpunished willful transgressions. The evaporation of plausible deniability.
I know braking data is used to identify dangerous road sites / locations, and dangerous prior driving behavior of this car. The dangerous road site versus driver can be disentangled by statistics: if the road site / location results in similar braking behavior in other drivers, its more associated to the site, if the braking behavior is more correlated with the driver, its more attributable to this driver. However most people tend to have relatively regular commutes due to location of their home, their job, and their working hours, their shopping patterns etc. so it still entangled with other drivers, since they will tend to encounter the same subset of drivers, also having their own relatively steady probabilistic patterns.
For example, when a user suddenly brakes with large delta v, is it really due to this driver's aptitude to not predict the results of their driving decisions? Or is it because they frequently encounter the same reckless drivers?
It seems this could also be detected: for each braking event, consider a disc of sufficient radius and similarily downscore other drivers in this disc, use proper Bayesian inference of course, not naive linear score incrementing decrementing...
Simply downrating the driver of the braking vehicle risks taxing the less reckless chickens vis-a-vis the dare's in chicken or dare scenario's, naive calculations risk taxing specifically those parties that decrease the total kinetic energy in potentially dangerous situations, if the reckless drivers don't flinch even if it would have gotten them into trouble if a chicken had been a reckless dare.
The insurance company doesn’t care whether the risk is because of the driver or because of the road that the driver is on. They are on the hook for the risk either way.
This is why insurance companies also use your zip code to rate you. If you live near roads with more losses, you are more likely to incur losses. Doesn’t matter if you’re a great driver or not — someone might hit you.
> When asked about the Pi Zero 2W, Eben said the substrate supply is constrained—basically, so many AI chips are being made that even older chips using older process nodes have to fight for the actual silicon wafers to use to make the chips.
I keep hearing voices invalidate each other, is the bottleneck the raw silicon substrate, or fab capacity?
What purity levels are required for say Pi Zero 2W?
The volume of monocrystalline silicon used in solar panels is orders of magnitude greater than the volume used in IC's / RAM production.
Is the actual bottleneck 11N + grade silicon wafers while 6N to 9N grade used in solar panels remains unaffected?
The teacher distillation is a corpus of text, and the "next token after the context" would be looking-up the context in the corpus, and for each occurrence the label is what followed in the corpus, scaled down by the number of occurrences of the context. The teacher is moot on contexts outside of the corpus though, unlike the usual teacher model in distillation.
a lot of the schematics have avoidable edge crossings, that could improve intuitive readability enormously, theres entire subfields of graph theory that consideres rendering of graphs and planar embeddings.
reply