881–890 of 978 posts
The lesson of the last 50 years of the computer and software marketplace is that free and low-end eventually wins. - PCs destroyed minicomputers. Mainframes survive, but serving a much tinier portion of the market than they used to. - PC office productivity software destroyed expensive professional products. - Windows (low end) and Linux (free) completely destroyed the UNIX marketplace, and again, have taken huge mar…
Earlier quoted context omitted.
> god-like So would you say we are months away from full self-driving cars that can out-drive a human being in any situation?
Remember, these cars are running local LLMs, not frontier models. The issue with self-driving cars has been the edge cases. The 0.0001% of situations where the models did not have sufficient training data. This is compounded by the hardware limitations. Onboard RAM in a typical Tesla on the road is 16GB (+16GB for the backup computer). This has to run the existing onboard OS and other operations plus the LLM. These t…
Earlier quoted context omitted.
Not sure what your point is. I’m commenting on a moment in time which is like an earlier moment in time. A different moment in time will have different analogs.
> The lesson of the last 50 years of the computer and software marketplace is that free and low-end eventually wins. Mac has won and it is not free. Iphone has won and it is not low end.
Earlier quoted context omitted.
> I don’t think the general-purpose LLM as a standalone tool is long for this world, at least not in consumer-facing applications. I use AI chat every day, I find it endlessly useful. It’s replaced google search.
> I use AI chat every day, I find it endlessly useful. It’s replaced google search. Extremely subsidized agentic search is very superior to Google at the moment, and of course it is. Google is a public company. The AI summary model has to work instantly, is likely as dumb as a 8T param model, and gives you incorrect details constantly. This sucks so much for Google. If you click on "AI Mode," suddenly the facts becom…
I’m into Claude for $20/mo (petty bourgeois tier), let’s assume for sake of examination this is only paying for power, infrastructure already amortized
Price of grid power approx $0.15/kWh
$20 / ($0.15/kWh * 30 days) = 4.4 kWh per day.
This same amount of energy can lift a 2-ton SUV 1/2 mile into the air. To me this seems astounding.
If only half the $20 pays for power it’s still quite impressive.
Earlier quoted context omitted.
Two wrongs don't make a right. Even if a certain large Asian country has carefully constructed a pretext to do do out of confected historical grievance, and entitlement to 'rise' at the expense of others?
First, copying information isn't wrong to begin with. It is literally the one thing that makes our species special. Second, even if you are a copyright maximalist the output of an LLM is either a) not subject to copyright because it is not the creative work of a human or b) a derivative work of the original training material to which the LLM's operator has no rights. Since the LLM's operator forcefully asserts that i…
Earlier quoted context omitted.
Personally, I don’t think the general-purpose LLM as a standalone tool is long for this world, at least not in consumer-facing applications. I think when the economics make more sense, product designers will make things that people actually want to use that will pretty transparently handle whatever model interactions are necessary, when it makes sense. As a consumer, the last things I want in an interface are to a) b…
>As a consumer, the last things I want in an interface are to a) be sycophantic enough to lessen my judgment This is EXACTLY what people like/are addicted to about chatbots. My sister-in-law bombed an interview and asked AI about her answers to the interviewer's questions, chatgpt or whatever it was told her that her answers weren't bad, but that the interviewer could not see the gold in her responses. She said she f…
> Chat rules : no sycophancy or over-agreeableness
(But even with that rule it's still necessary to be discerning about the responses you get and to push back against points made, or words used)
Earlier quoted context omitted.
> Put all of these trends together, and I think that in 10-15 years, we are going to have consumer PCs (and phones!) running models doing pretty much anything that frontier models can do right now. Phones are constrained by battery power and memory does not shrink as fast as CPU/GPU, so unless there's a battery breakthrough and/or memory breakthrough, you're not fitting 100Gb of RAM on your phone in 10 years. Absolut…
Not just that memory shrinks slower, it has practically completely stalled. On chip cache seems stuck at 7nm and DRAM is stuck at 10nm. As transistors shrink, they hold less charge, creating weaker signals that are harder to read and prone to interference. Smaller nodes aren't a huge issue on CPU/GPU work load because they don't have to hold a static state. I'm not saying we are at peak memory but future gains are go…
The lesson of the last 50 years of the computer and software marketplace is that free and low-end eventually wins. - PCs destroyed minicomputers. Mainframes survive, but serving a much tinier portion of the market than they used to. - PC office productivity software destroyed expensive professional products. - Windows (low end) and Linux (free) completely destroyed the UNIX marketplace, and again, have taken huge mar…
The problem (right now) is that Open Weight models depend right now on huge companies to spend billion of dollars to train and develop them, all backed up by their incentives and their state to support this, while essentially giving away their monetization path. With open source projects, the benefit was that each individual could improve the complex system (e.g. Linux Kernel) interpedently, and over time the benefit…
Earlier quoted context omitted.
The problem (right now) is that Open Weight models depend right now on huge companies to spend billion of dollars to train and develop them, all backed up by their incentives and their state to support this, while essentially giving away their monetization path. With open source projects, the benefit was that each individual could improve the complex system (e.g. Linux Kernel) interpedently, and over time the benefit…
This is precisely why we need more projects like this https://github.com/bigscience-workshop/petals There's no fundamental reason why models couldn't be developed and trained using community efforts. It might not be as fast and efficient, but it's definitely possible.
The lesson of the last 50 years of the computer and software marketplace is that free and low-end eventually wins. - PCs destroyed minicomputers. Mainframes survive, but serving a much tinier portion of the market than they used to. - PC office productivity software destroyed expensive professional products. - Windows (low end) and Linux (free) completely destroyed the UNIX marketplace, and again, have taken huge mar…
Another important thing that made software usage and education available for most of the world was piracy. I remember as a kid growing up in a developing country, any software (windows, office, Visual Basic, flash, dreamweaver, etc.) was less than 1$. That allowed me to try out and learn so many things on my own without paying a huge amount of money for the license. And I think this is true for most of the software d…
This is all (Microsoft) junk and so I wonder if you actually benefitted from this 'piracy'. And of course, it's well known that MS turned a blind eye to such 'piracy' in lesser developed countries, as they knew that they were gaining a future paying customer base.