"VLM^3: VLMs Are Native 3D Learners" This paper shows that VLMs can learn 3D natively. Most 3D vision systems rely on expert architectures, regression heads, heavy augmentations, and task-specific losses. But they show that you can skip the majority of these designs. All they
RESEARCH
-
AI identifies cat; bounding box pixel decoding is slow
By
–
An AI can tell you there's a cat in the image. Pointing to the exact pixels is the hard part.
— Satya Mallick (@LearnOpenCV) 2 juin 2026
The reason it's slow: most VLMs spell out a bounding box one coordinate token at a time — some even split "1024" into single digits. But a box's corners are connected. Decode them… pic.twitter.com/eoJu0PiHGUAn AI can tell you there's a cat in the image. Pointing to the exact pixels is the hard part.
The reason it's slow: most VLMs spell out a bounding box one coordinate token at a time — some even split "1024" into single digits. But a box's corners are connected. Decode them -
Mustafa Suleyman predicts three OOM jumps in training compute by 2029
By
–
https://t.co/LizAzIh7dN pic.twitter.com/VVreg7R7OW
— Bojan Tunguz (@tunguz) 2 juin 2026At Microsoft Build today, Mustafa Suleyman predicted three more OOM jumps in the amount of training compute between now and summer 2029.
-
Benchtalks #2: SWE-bench creator jyangballin discusses ProgramBench and upcoming project
By
–
Benchtalks #2: @jyangballin
, creator of SWE-bench, fresh off ProgramBench (every frontier model: 0% at launch). Out soon with @vincentsunnchen
. -
Comparison of Opus/sonnet to sonnet 4.6 on SWE pro
By
–
Re Opus/sonnet: from what I understood they compare it to sonnet 4.6. only on SWE pro comparable to opus. If I’m correct the quote was „side by side with sonnet 4.6“
-
AI crunches SDR and real estate manuals to understand human psychology
By
–
Human psychology is a complex thing. It can be actively reasoned about. We now have surprisingly powerful reasoning machines. They have crunched through many sources of data, including SaaS SDR training manuals and real estate agent how-to books.
-
Microsoft AI technical write-up praised for rare detail and quality
By
–
Kudos to the team on a really good technical write-up, though, which a lot of AI labs are no longer providing. https://
microsoft.ai/wp-content/upl
oads/2026/06/main_20260602_2.pdf
… -
MAI-Thinking-1 hard to evaluate, stats below Meta Spark
By
–
It is difficult to know how good MAI-Thinking-1 is from the scores alone (like weirdly low GPQA & Terminal Bench 2.0) But Microsoft makes it really hard to try its models upon release (a general issue with many Microsoft AI products), so I dunno. Stats below Meta Spark, though.
-
NeurIPS position paper track overwhelmed by AI submissions, chairs take action
By
–
The NeurIPS position paper track was flooded with submissions that substantially used AI. More than other tracks, and despite a requirement otherwise. I appreciate the chairs' strong action. People are not entitled to reviewers' time, AI makes it exceptionally easy to waste it.
