It’s a new day at Digg. Fresh stories are clustering now. 90% complete.
Top Tech Stories
1
Top Story
Alibaba announces 2.4-trillion-parameter open-weight Qwen3.8 model
A preview version is available now for testing.
2
Dean Ball argues open-weight AI models function as market dumping
DeepMind's Andreas Kirsch warns compute providers capture the returns.
3
Altimeter's Brad Gerstner warns closed-source AI labs seek regulatory capture
Critics say labs lobby agencies to restrict open-weight models.
4
Scott Belsky and Pieter Levels track rise of temporary startups
Operators deploy temporary AI wrappers to capture brief arbitrage.
5
Vercel's Guillermo Rauch finds open-weight Kimi K3 achieves frontier-level cybersecurity
Sol performed better than Kimi K3 but cost much more.
6
Kalomaze compares Mythos looped transformer claims to OpenAI o1 rumors
The debate highlights recurring community speculation about proprietary model architectures
7
OpenAI's Dean Ball says open-weight releases slow capex, not progress
He faced public pushback after analyzing Chinese AI model Kimi.
8
Garry Tan advocates "file over app" philosophy over transient AI
Tan highlights Markdown as a resilient, universal data format.
9
GPT 5.6 resolves 70-year-old Erdős conjecture with one-page proof
The solution replaces a 44-page proof from 1991.
10
Harry Stebbings replaces Salesforce with AI CRM, saving $600,000
The transition sparked a debate on SaaS market survival.
11
Yacine, systems engineer, reports ChatGPT autonomously debugged his codebase
The agent optimized code during a 10-hour hyperparameter sweep.
12
Researchers defend open-source AI against dumping claims
They note proprietary labs rely on PyTorch and JAX.
13
Microsoft Research's Dimitris Papailiopoulos jokes about "tree-shade communism
Another user joked about building a floating datacenter there.
14
Tim Dettmers, bitsandbytes creator, finds loop-based AI agents unreliable
Isolated single-agent overnight sessions bypassed compounding drift.
15
Eli Dourado and Jason Lee request multi-account ChatGPT Gmail support
They want to manage separate work and personal inboxes.
16
Together AI teases K3 speed optimizations after slow testing feedback
Some users reported testing cycles taking multiple days.
17
Zephyr estimates GB300 NVL72 RL rollouts cost $3.5 million
Long-term agreements can cut assumed GPU rates in half.
18NEW
Hugging Face's Merve Noyan warns against restricting frontier AI training
Andreas Kirsch argues a monopoly could accelerate superintelligence.