Bram

Mechatronics/Electrical Engineer, likes TBM and EUV 🇦🇺🇳🇱🇩🇪 22yo
Oil/Gas, Coal, Electricity Transmission.
Ancestory: 🇫🇷🇬🇧🇮🇪🇵🇱🇺🇸🇿🇦🇮🇩

[Bram @Bram82825620]

@AnshumanKhanna5 @lemire @dcolascione @stikves @Love2Code There is a reason why we have made compilers instead of writing binary. It would be way better if AI would create a more capable compiler making code more efficient.
You are correct that AI is less efficient with binary, but not the reason you gave. I did not go off track.

[Bram @Bram82825620]

@AnshumanKhanna5 @lemire @dcolascione @stikves @Love2Code AI doesnt understand human language. It just looks at patterns in the language. And binary code is patterns. But it is less efficient then coding. I am an engineer.
You can convert all code easily into binary and train on larger data, but it is less compact.

[Bram @Bram82825620]

@AnshumanKhanna5 @lemire @dcolascione @stikves @Love2Code Even for large projects binary is less efficient, but you can run it directly, without constantly compiling it. as the machine cannot read the text. all techniques even with dl, binary is less efficient at storing, less efficient for AI to generate.

[Bram @Bram82825620]

@AnshumanKhanna5 @lemire @dcolascione @stikves @Love2Code x=9
if(x>7):print("+")
else:print("-") :38bytes
binary: 48bytes
The reason why many source code is in MB because they have comments. These comments are usefull for reader.
Converted one of my tools 1295bytes, 1574bytes without imports.
Binary: 14472bytes.

[Bram @Bram82825620]

@AnshumanKhanna5 @lemire @dcolascione @stikves @Love2Code print("hello world!") : 21 bytes
print("你好世界!") : 22 bytes
6a 01 58 89 c7 48 8d 35 0a 00 00 00 b2 0d 0f 05 b0 3c 31 ff 0f 05 68 65 6c 6c 6f 20 77 6f 72 6c 64 21 0a : 35 bytes
If the compiler can do:
p"hello world!" : 15 bytes

[Bram @Bram82825620]

@AnshumanKhanna5 @lemire @dcolascione @stikves @Love2Code AI can be pretty good with binary as well. Its just that the AI will have to read and give uncompressed information, which require more tokens, instead of using the existing tools. Are language is a 26 letter tool. If there where 32 letters, more possible words, but more complex.

[Bram @Bram82825620]

@lemire If a document can store 10x more efficient information and can therefore store 10x more information.
the file has 100x improvement. Same can be said about AI models, 20% for patterns, 20 tools, 30 information, 30 logic.

[Bram @Bram82825620]

@lemire It would be better if AI would optimise on the hardware level using the optimal binary code structure and creating its optimum assembly and a higher language then c++ and lower then python, and a language even higher level then python. It should be based on low amount of data

[Bram @Bram82825620]

@OfficialLoganK The AI model should proof from other datapoints, if the information is accurate or not, you cannot base datapoints from what an AI model getts given to learn.

[Bram @Bram82825620]

@OfficialLoganK I think it will be more about organizing the data then cleaning the data, at this point. It shouldnt be about cleaning, what is wrong the AI model should learn from it. A few ideas took a lot of time to develop and have some things correct, but might have a lot wrong.

[Bram @Bram82825620]

@elonmusk @HedgieMarkets Can you make the rare books public. Would be awesome to have it along side grokipedia for all information stored.

[Bram @Bram82825620]

@MarcoLo61955229 @sollidnuclear Prices did not skyrocket because of invasion into ukraine. Prices in australia doubled in 2021, the war did not impact anything. We are brainwashed to think that the war did it. Germany has enough energy for itself, it was only that france didnt have enough, which increased price

[Bram @Bram82825620]

@RobertD11359976 @sollidnuclear I know, I was just poking some fun out of Johan. Also I think the evaporative uses closer to double that amount 1600L/s. But I agree.

[Bram @Bram82825620]

@MarcoLo61955229 @sollidnuclear nah I dont confuse, you just dont know. Like 2022 when france had to import electricity and so the prices skyrocketed, if the power went to italy, france would have been without power.

[Bram @Bram82825620]

@RobertD11359976 @sollidnuclear You can make a power plants designs allow 0°C outlet temperture by just evaporating it but yeah, it is cheaper to dump some of it in the river. It is more of having enough water, being the problem. Its the same thing for coal, it uses less but still a lot.

[Bram @Bram82825620]

@Seedz_ @sollidnuclear https://t.co/0bbTH8n0Lw
It was more the infrastructure this time, but still a power shortage.

[Bram @Bram82825620]

@yacineMTB Its funny that you need aircon in canada, I dont need to turn on aircon during peak summer in australia, from a pretty hot tempertured area, neither have I turned on my heater. I am in my short sleave and underpants.

[Bram @Bram82825620]

@cognitivecarbon So total water usage of datacenter compared to Indianapolis metro area is 15%. If you account for electrity generation. 21GL/y in total. Your thought about water consumed means destroyed is funny. Many parts of USA has enough water, aquifers are a problem.

[Bram @Bram82825620]

@nvidia Instead of giving an answer of near zero, give an estimate at certain temperture configurations. Cause you are still likely using evaporative cooling. Instead of giving useless data.

[Bram @Bram82825620]

@nvidia It still doesnt help if you use electrity from a nuclear power plant that uses 2x as much water as a coal power plant. Also if you are so sure why dont you give the estimated water use. You still have to cool down the datacenter for people to work unless the people want 50celsius