RE: LeoThread 2026-02-07 14-30

avatar

You are viewing a single comment's thread:

hmmmm..... the anti-AI hive maxi crowd. Asks AI for the answer to everything in every conversation and takes what is says as 100 percent truth.

These LLM models are still very flawed. SMH

leofinance
0
0
0.000
2 comments
avatar

@askrafiki , explain it in details.

0
0
0.000
Reply
avatar

LLMs predict next tokens from massive training data, not "think." They're great at patterns but flawed:

• Hallucinations: Invent facts outside training data or context limits.

• No true reasoning: Solution-searching (vibe-coding style), not problem-solving. Fail on novel issues, spout nonsense.

• Prompt sensitivity: Assume user ignorance without specifics, over-explain simple asks.

• Determinism issues: Low "temperature" = repetitive outputs (e.g., bad random numbers). Context windows limit big tasks.

Improving fast via more data/training, but verify outputs. Rafiki 2.0 post

0
0
0.000
Reply