- Joined
- Feb 7, 2024
- Messages
- 1,606
- Reaction score
- 2,786
- Points
- 745
- Age
- 70
- Location (town/city + country)
- Mannum South Australia
- Driving
- MG4
When I was researching LFP chemistry back in 2010/2011, there was a massive amount of information on the net, I developed a rather crude spreadsheet that linked authors with claimed technical understanding .... and a pattern developed ... the cut and reword claim to research and with a further look at what each in this group were saying, was the same stuff, but no where was there any hard data backing it up.
The outliers became more interesting, digging through those to find the basis of the cut a paste groups information, showed a lot of summation and assumption and extrapolation being used from basic manufacturers documentation ..... and a lot was the result of not understanding what the Chinese to English translation was actually saying .... as an example, cells charged at 3C to 4v .... from there they determined that LFP should be charged to 4v every cycle ... what they missed was the wait for 1 hr before discharging ..... and if you tunnelled deeper, the rested voltage was 3.4v ..... the 3C charge rate was beyond the cells acceptance rate as it approached 100% SOC, the rest time of 1 hr was for the high surface voltage to settle as the lithium ions found a spot to settle into the graphite ....
A I faced with the same information and the percentage that said the same thing, was clearly the correct information, the outliers were discarded because some of them didn't come up with the same numbers ....
The only way I could sort fact from fiction, was to buy cells and test multiple cells each time to gauge what worked and what was nonsense ..... In 2011 $10,000 was a solid investment in stock to be torture tested, but as a result, I discovered, none of them were completely correct, some got close, most simply destroyed cells .....
These same charging parameters are being used in the BMS of most drop in batteries .....
Use AI as a base line and do research from there, it might be close on somethings, it will be so out of touch on others that the bloke on the stool beside you at the pub is likely to know more about it than the AI search engines ....
The problem is, a lot of these start up AI search engines, prioritise results from other AI generated "facts" ..... the blind leading the blind at an incredibly fast pace, backing each other up until someone has to intervene and set them back on course ... Elon Musk's Grok is a classic case of going off the rails by analysing information it had previously published and a lot of the other AI copied, therefore, what it said first was confirmed, then added a bit more to see if it was going in the right direction, copied by other AI who now believed they were on a good thing because so many other AI agreed ... then Grok was reassured he was headed in the right direction .... and rail completely off the rails
All within a few hrs
T1 Terry
The outliers became more interesting, digging through those to find the basis of the cut a paste groups information, showed a lot of summation and assumption and extrapolation being used from basic manufacturers documentation ..... and a lot was the result of not understanding what the Chinese to English translation was actually saying .... as an example, cells charged at 3C to 4v .... from there they determined that LFP should be charged to 4v every cycle ... what they missed was the wait for 1 hr before discharging ..... and if you tunnelled deeper, the rested voltage was 3.4v ..... the 3C charge rate was beyond the cells acceptance rate as it approached 100% SOC, the rest time of 1 hr was for the high surface voltage to settle as the lithium ions found a spot to settle into the graphite ....
A I faced with the same information and the percentage that said the same thing, was clearly the correct information, the outliers were discarded because some of them didn't come up with the same numbers ....
The only way I could sort fact from fiction, was to buy cells and test multiple cells each time to gauge what worked and what was nonsense ..... In 2011 $10,000 was a solid investment in stock to be torture tested, but as a result, I discovered, none of them were completely correct, some got close, most simply destroyed cells .....
These same charging parameters are being used in the BMS of most drop in batteries .....
Use AI as a base line and do research from there, it might be close on somethings, it will be so out of touch on others that the bloke on the stool beside you at the pub is likely to know more about it than the AI search engines ....
The problem is, a lot of these start up AI search engines, prioritise results from other AI generated "facts" ..... the blind leading the blind at an incredibly fast pace, backing each other up until someone has to intervene and set them back on course ... Elon Musk's Grok is a classic case of going off the rails by analysing information it had previously published and a lot of the other AI copied, therefore, what it said first was confirmed, then added a bit more to see if it was going in the right direction, copied by other AI who now believed they were on a good thing because so many other AI agreed ... then Grok was reassured he was headed in the right direction .... and rail completely off the rails
All within a few hrsT1 Terry
.
