Question. How does it work if I own a repository (opt out, don't use copilot) and I give access to someone else (use is opted in and uses copilot).
Do you train on his submissions of my code?
How can you know what that he has the right to share the code with you for training?
You might want your browser to do Bluetooth, NFC, Background stuff, Face Detection but I don't.
I like to use Apple products for things that are commodities to me because I am not gonna look into the details of those and when I do Apple reasoning often make sense to me (just like this list).
There is a lot more we can criticize about these big tech corps (including Apple) than a product decision for a company that is known for making polarizing decisions on behalf of their customers. If people buy it... they must like it, no?
exactly what I do. they are still annoying because providers want to me to be engaged on their terms and keep pushing these features and don’t offer a broad “never use bluetooth on any website ever” kind of option. Magic of KPIs
anyway the point here is that I don’t really care if Safari is behind in support. The article was about blaming Safari for being behind.
My question is: is safer than average human good enough?
When I drive I have the option to choose to be safe or not. When a computer drives I lose that option. So for 49% of the people, safer than the average human is less safe than before.
I think we need to reach "Safer than the safest 10% of humans".
Also these reports should be done by a government agency.
Yes, it's good enough. Because you cannot control who else is on the street around you. Having cars around you that are driving safer than the average is better than them driving average.
You seem to be thinking this is compared to median driver, and not average one: if most accidents are caused by small number of bad drivers, your average will be bad, but even median might still be good.
And this does not even compare the drivers, but simply miles driven.
So I think that 80% of human drivers would likely be safer than Waymo unless they are driving under the influence or extremely tired or distracted.
Note that "13x safer" already implies being in the top 10%, though.
From a purely technical perspective, I would expect a Waymo to react to unexpected external stimuli much, much faster than a human. It gives the Waymo options that are unavailable to a human's slower reaction time.
Hi! Yes, the premise is exactly to review how 'stable' agents would be if left unsupervised in an end-2-end scenario.
But this is probably unrealistic now, hence the experiment. I think people will be less skeptical the more they interact with these kind of entities and slowly develop trust.
That's why we developed agents with an identity and primarily around email in order to 'plug' them into company processes slowly and naturally. That's the core idea of the main project this experiment spawn off of.
I’m running a public experiment in multi-agent coordination: can a small team of independent AI employees collaborate toward a shared objective using the same tools regular companies already use?
In this experiment, each AI employee has a specific identity (role, personality, scope, and permissions) and their own email address and access to corporate tools.
They coordinate primarily over email and a shared Google Sheet, plus they connect to the paper brokerage account (I use Alpaca Markets).
Platypi Capital is the “sandbox”: the team runs a paper trading portfolio. The employees research, debate, propose trades and strategies, do risk checks, and execute trades (paper money) as a coordinated workflow, then publish positions/orders/performance. This is completely transparent and realtime on the website.
This is not a real fund. This is an experiment on how AIs coordinate together. Trades are executed with paper money on a simulated brokerage account, and nothing here is financial advice. This is part of a broader effort I’m working on to build an “AI employees” product.
It's interesting how different dishwashers are in US and Europe. Two main things for me:
- Salt: European dishwashers have embedded water softeners and you add salt once in a while. Only super high end ones have it in US.
- Water heater: European dishwasher expect to receive cold water and they heat it internally; US ones expect hot water and only partially boost the temperature (sometimes). That's why you have to run hot water before starting the dishwasher
Yes, but you need energy to pump heat, and that has an efficiency maximum (thx ~~Obama~~ Carnot), and radiative cooling scales with the ~4th power of the temperature, so it has to be really hot, and so it requires a lot of energy to "cool down" the already relatively cool side and use that "heat" to heat up the other side that's a thousand degree hotter.
All in all, the cooling system would likely consume more energy than the compute parts.
yes. it is how sats currently handle this. its actually exponentially effective too P = E S A T^4
requires a lot of weight (cooling fluid). requires a lot of materials science (dont want to burn out radiator). requires a lot of moving parts (sun shutters if your orbit ever faces the sun - radiator is going to be both ways).
so that sounds all well and good (wow! 4th power efficiency!) but it's still insanely expensive and if your radiator solution fucks up in any way (in famously easy to service environment space) then your entire investment is toast
now i havent run the math on cost or what elon thinks the cost is, but my extremely favorable back of hand math suggests he's full of it
Be careful with the math there. While a 4th power is awesome you got the Stefan-Boltzman constant to consider and that's on the order of 10^-8
Radiative power is really efficient for hot things but not so great when you're trying to keep things down to normal levels. Efficient for shedding heat from a sun but not so much for keeping a cpu from overheating...
You can. This is how it is currently done, but it is not easy. It needs to have a large enough surface area to radiate the heat, and also be protected from the sun (as to not collect extra heat). For a data centre, think of an at least 1000m2 heat exchange panel (likely more to train a frontier model).
You can, but the heat needs to go somewhere, and now you're back to square one, with "how do I get rid of all this heat". Earth refrigerators have a large heat exchanger on the back for this purpose. In fact now you need to get rid of both of the heat your compute generates and the energy your refrigerator pump uses - an example people often give is that a fridge with an open door actually heats the room, as it spends energy on moving heat around pointlessly.
You definitely _can_ the question is, can you do it by enough for a reasonable amount of money. There are a few techniques to this but at the end of the day you need to radiate away, the heat otherwise it will just keep growing. You cannot keep pumping energy into the satellite without distributing the same amount back out again.
I don't think that kind of difference in benchmarks has any meaning at all. Your agentic coding tool and the task you are working on introduce a lot more "noise" than that small delta.
Also consider they are all overfitting on the benchmark itself so there might be that as well (which can go in either directions)
I consider the top models practically identical for coding applications (just personal experience with heavy use of both GPT5.2 and Opus 4.5).
Excited to see how this model compares in real applications. It's 1/5th of the price of top models!!
