Latest

Every published item, in order · 91 items
10月13日
ModelsBuildersMulti-sourceSelected13:01
MiniCPM-V 4.7 runs 35B multimodal model on phones

The key change is deployment economics: a 35B MoE now fits consumer devices at usable speeds, because LES hides IO wait through scheduling instead of trading off precision. Developers building privacy-sensitive on-device apps should test it, since no cloud call is required. Against typical acceleration methods that cut quantisation levels and lose points, this keeps baseline accuracy intact.

Event
AIGCLINK@aigclink6 sourcesSource
ModelsBuildersSelected03:56
Meta tests Muse Wonder Full scaling on unfamiliar tasks

The finding suggests base model quality, not more post-training, drives generalisation to tasks outside the training distribution — a useful signal for teams deciding where to spend compute. Researchers evaluating reasoning models can use this framing to compare candidates on out-of-distribution scaling curves rather than benchmark scores alone. No regional angle exists; the most concrete comparison is Muse Wonder Full against the Muse Spark 1.3 trend line.

官方账号Meta AI@AIatMetaSource
ProductsBuildersSelected03:26
Meta Muse Wonder Adds Active Memory and Active Focus for Agents

Active Memory and Active Focus address the two most common failure points of agents today: losing context between sessions and stalling mid-task. Teams building or selecting agent tools should compare Muse Wonder's persistence features against alternatives like Claude or ChatGPT agents when evaluating multi-session workflows. No pricing or regional availability details were given in the announcement.

官方账号Meta AI@AIatMetaSource