AI Training
6 stories in this blend
Superhuman·2d agoTech companies compete in bankruptcy auction for Spirit Airlines' workplace records
Artificial intelligence developers are bidding millions of dollars to acquire internal communications and operational archives from the defunct carrier Spirit Airlines. Tech firms plan to use the hundreds of millions of corporate emails and chat records to train models on real-world organizational tasks.
Superhuman·5d agoAnthropic launches educational platform and prepares possible stock offering
Anthropic has opened a centralized repository called The Claude Academy featuring over 300 tutorials and guides across its software suite. The company also introduced a concise response setting for developer workflows while reportedly preparing initial documentation to go public.
AI Daily Brief·Aug 20Google wins bankruptcy auction for corporate internal communications data
Google outbid competitor Mercor to purchase Spirit Airlines' internal corporate records for ten million dollars in a bankruptcy proceeding. The deal excludes consumer credit details and personal files, focusing instead on decades of employee emails, internal messaging logs, and meeting transcripts. Technology companies are increasingly purchasing real-world business records to train automated workplace agents on actual corporate operations.
Superintelligence·Aug 19OpenAI Halts Major Model Training Run Over Alignment Concerns
OpenAI temporarily suspended reinforcement learning on its upcoming deployment models and put its largest planned training run on hold. The organization cited internal cybersecurity risk triggers and stated it will pause until smaller test runs demonstrate adequate alignment controls.
- SuperintelligenceAI Training
Z.ai boosts model performance through targeted post-training instead of building larger systems
Superintelligence·Aug 15Z.ai boosts model performance through targeted post-training instead of building larger systems
Chinese artificial intelligence lab Z.ai launched GLM-5.3 using the identical 743-billion parameter foundational architecture as its prior version, relying entirely on post-training refinements. The model's score on the command-line Terminal-Bench 3.0 benchmark jumped from 4.6 to 28.3 within 59 days. However, performance improvements across other general testing categories remained substantially smaller, showing uneven gains.
Superintelligence·Aug 12Google Trains Gemini AI Model Using Simulated Medical Encounters
Google and DeepMind published results on ResidencyRL, a system that trained Gemini 3.5 Flash through nearly 50,000 simulated doctor-patient interactions. By practicing diagnostic routines and communication skills against automated patient personas, the model improved its clinical accuracy from 81% to 88%.
