Wow! Just tested EXAONE Deep (2.4B, 7.8B, 32B) and I'm impressed. It's really fast and great at reasoning: Even the 2.4B model solved "Count to ten, skipping numbers ending in 'e'" right away, which all ≤32B models I tested before failed. It does seem to hallucinate sometimes, but its reasoning is amazing!

>>> Count from 0 to ten while skipping every number ending in 'e'.

<thought>
[...]
</thought>
To solve the problem of counting from 0 to ten while skipping numbers whose spelled-out form ends with the letter 'e',
we first list each number from 0 to 10 and check the last letter of their English spellings:

- **0 (zero)**: Ends with 'O' → keep
- **1 (one)**: Ends with 'E' → skip
- **2 (two)**: Ends with 'O' → keep
- **3 (three)**: Ends with 'E' → skip
- **4 (four)**: Ends with 'R' → keep
- **5 (five)**: Ends with 'E' → skip
- **6 (six)**: Ends with 'X' → keep
- **7 (seven)**: Ends with 'M' → keep
- **8 (eight)**: Ends with 'T' → keep
- **9 (nine)**: Ends with 'E' → skip
- **10 (ten)**: Ends with 'N' → keep

The numbers retained are: 0, 2, 4, 6, 7, 8, and 10.

#exaonedeep #exaone #ai

Recent searches

Search options

Administered by:

Server stats:

#impressive