←Search results

Search “Reasoning”

By relevance · Total 3279 results Clear search

Monday

  1. Exponential View

    Exponential View Data Weekly: Open-weight token share doubles in a year, AI agent token usage reaches 14 times that of humans

    The latest inference token data from Exponential View shows that the open-weight token share has doubled over the past twelve months and is approaching a 1:1 ratio of closed-source to open-source, but over the same period the number of closed-weight tokens grew sevenfold. The volume of tokens used by AI agents in February of this year has already surpassed that of humans, and currently stands at 14 times that of humans, while human token usage grew only 2.8 times. The employment rate of young workers aged 22–25 in AI-exposed occupations is now 19% below the trend line, up from 15% last year.

    The latest inference token data from Exponential View shows that the open-weight token share has doubled over the past twelve months and is approaching a 1:1 ratio of closed-source to open-source, but over the same period the number of closed-weight tokens grew sevenfold. The volume of tokens used by AI agents in February of this year has already surpassed that of humans, and currently stands at 14 times that of humans, while human token usage grew only 2.8 times. The employment rate of young wo

    Matched fieldsSummary、Body

    #newsletter#English text is being prepared.#English text is being prepared.#Agents

Saturday

Friday

Wednesday

Tuesday

Monday

Friday

  1. Ant Ling: Developer Blog

    Ant Ling closes the single-machine Agentic RL loop with Ling-3.0-tiny and AReno

    Ant Ling has published a practice article describing how it used the local post-training toolkit AReno on a DGX Spark to perform Agentic RL on Ling-3.0-tiny, using tic-tac-toe as a minimal validation task to complete the four-step loop of baseline evaluation, Reward definition, post-training, and re-evaluation in the same environment. The experiment used GSPO for 400 training steps; rollout/rewards_mean rose from about -0.5 to around 0.4, response_len dropped to about 850 tokens, and tool calling and move selection became more stable in the re-evaluation. Ling-3.0-tiny is a hybrid reasoning model with 7.9B total parameters and 1.3B activated parameters; BF16, FP8, and INT4 versions have been open-sourced, and AReno provides a Docker reproduction tutorial.

    English text is being prepared.

    Ant Ling has published a practice article describing how it used the local post-training toolkit AReno on a DGX Spark to perform Agentic RL on Ling-3.0-tiny, using tic-tac-toe as a minimal validation task to complete the four-step loop of baseline evaluation, Reward definition, post-training, and re-evaluation in the same environment. The experiment used GSPO for 400 training steps; rollout/rewards mean rose from about -0.5 to around 0.4, response len dropped to about 850 tokens, and tool callin

    Matched fieldsSummary、Body

    #engineering#Tutorials / practice#Agents#Open-source ecosystem

Thursday

Wednesday