Olle
Region: CH
Thursday 18 June 2026 17:47:56 GMT
8735
371
13
47
Music
Download
Comments
Wolf :
Or in other words, never trust benchmarks (alone) models get trained specifically to score high on public benchmarks. That does not mean they are accually good. It's a small part that does help them, but also means they lack in all other departments. 🙃 À 1b parameter model can out perform 1T parameter models in benchmarks. Does not mean that 1b model is better 😀
2026-06-20 08:44:51
15
Newcomer1989 :
anyone tested it? is it good with simple tool calls? how long is the context?
2026-07-02 21:37:21
0
Free AI Money :
Finally someone else looking at other llms!
2026-06-18 19:41:07
2
El Consulul :
I tried it and it's insanely good for its size
2026-06-25 15:50:10
1
Brave Guido :
Are you telling me they just made a perfect 1 tick predictor engine
2026-06-19 20:25:40
0
Raj—> :
They optimize these tiny models for benchmarking that doesn't make them better. You could argue that makes some worse.
2026-06-25 14:41:21
0
DataAnalystNi :
Nice. Can you please correct me if I understood wrong, I think Andrej Karpathy said that models are specialized in certain things and not improving on others, I.e. tell jokes, specially i think for the amount of fine tunning required is that correct is that why Chinese models are good in coding but bad at reasoning or intelligence. Also is there something like this model you show specialized in coding?
2026-06-18 17:59:59
2
To see more videos from user @olleai, please go to the Tikwm
homepage.
