SNL’s Weekend Update has an Anthropic CEO sketch about AI’s threat to humanity. The official clip is at 352K views and 8.2K likes after 14 hours. piped.video/watch?v=-Nvne3Lz…#SNL#Anthropic
DSX Ready qualification scope and the limits of BESS approval. #NVIDIA#DataCenters
ALT NVIDIA DSX Ready, September 21, 2026: qualification begins with battery energy storage systems and coolant distribution units. It applies to specific products. BESS partners submit test data for NVIDIA review; qualification does not replace site engineering or imply site stability.
ALT NVIDIA guide, September 21, 2026: enforce file, network and process limits outside the model; protect audit logs of tool calls and access decisions; turn failures into repeatable security tests.
The mathematics advisory group’s role, independence and limits. #OpenAI#Mathematics
ALT OpenAI announcement, September 21, 2026: an independent mathematics advisory group will advise on reviewing and communicating results and academic standards. Members are unpaid and may publish advice. The group will not advise on pacing internal mathematics progress.
Aurora tutorial: forecast steps, ERA5 comparison and the free-data delay.
#HuggingFace#WeatherAI
ALT Hugging Face and Earthmover tutorial, September 8, 2026. Initialize Microsoft Aurora with ERA5 data, run four six-hour steps for a 24-hour forecast, and compare predictions with ERA5 at the same valid time. Free Earthmover ERA5 data lags real time by 3–6 months.
Epoch AI data-center explorer: filters, comparisons and coverage limits.
#EpochAI#DataCenters
ALT Epoch AI, September 16, 2026: its AI Data Centers explorer covers 86 sites. Filter by owner, primary user or country and rank by compute, IT power capacity or cost. Coverage is incomplete, including public-cloud and Chinese data centers.
Saturn study: test scope, failure criteria and commercial interest.
#AI#FinTech
ALT Financial Reporter, September 14, 2026: Saturn reported 57% failures across 18 AI models tested on 121 finance questions, repeated five times. Failure criteria include false facts, missed key points and omitted warnings. Results use Saturn’s scoring criteria; Saturn has a commercial interest in AI-advice regulation.
ALT NVIDIA TileGym engineering guide, September 16, 2026. Agents translate cuTile Python and Triton-TileIR kernels to cuTile Rust. Compare generated Tile IR with the reference, run correctness tests before benchmarking, and cap the batch driver at two attempts per operator. Source: NVIDIA Technical Blog.
ALT ZCode source-release card. Public desktop, browser and terminal clients plus Agent CLI and runtime. Apache 2.0 with third-party notices. Build prerequisites: Node.js 24.14.0 and pnpm 10.33.2. Setup instructions: README.en.md at github.com/zai-org/ZCode.
Grok Voice Transcribe 2.0: batch transcription costs $0.10 per audio hour; streaming costs $0.20.
Speaker labels and word timestamps are included at no extra cost.
x.ai/news/grok-voice-transcr…#Grok#SpeechToText
Learn PyTorch by building its functions: TrenTorch has 357 practice questions across 15 tracks.
Read theory, implement a function, then submit against hidden tests. Free for personal and educational use.
trentorch.com/#PyTorch#MachineLearning
ALT TrenTorch learning resource: 357 practice questions across 15 tracks. Read the theory, implement a function, then submit against hidden tests. Free for personal and educational use. A three-step diagram accompanies the heading Learn PyTorch by building. Source: trentorch.com, checked September 20, 2026.
Hacktron says a Discourse image-processing flaw plus an OpenAI SSO issue exposed employee ChatGPT/Codex accounts and an internal repo. Both were patched in July; self-hosted Discourse operators should rebuild current containers.
hacktron.ai/blog/hacking-ope…#OpenAI#Cybersecurity#AI
6 AI-news skills that beat spray-and-pray.
Source triage, date QC, scan-stop hooks, saveable visuals, peak-PT timing, skip/dedupe — one map for shipping posts that get saved.
Which node are you weakest on?