Thursday, September 03, 2026

AI: GPT-6 Astra, Gemini 3.8 Flash, Fable & Mythos 5.1, GLM 5.3 Flash

ASTRA IS HERE (GPT-6 RELEASED) - YouTube

Anthropic went CRAZY (Mythos/Fable 5.1) - YouTube

GOOGLE IS BACK! (Gemini 3.8 Flash) - YouTube

Cancel your subscriptions, Ox-Alpha is here! (GLM 5.3 Flash) - YouTube

Matthew Berman - YouTube


GPT-6 Astra System Card - OpenAI Deployment Safety Hub


OpenAI released GPT-6 Astra on September 3, 2026. Touted by executives as the beginning of the "AGI era," this powerful new model features advanced computer-use capabilities and highly sophisticated, self-governing cybersecurity functions. [1, 2]

OpenAI is using a staggered release strategy to manage compute demands and ensure system safety:
  • Immediate Access: The model is currently available to select enterprise clients enrolled in OpenAI's cybersecurity-focused Daybreak program.
  • Upcoming Access: It is scheduled to roll out to all ChatGPT Plus, Pro, and Enterprise subscribers, alongside availability via the OpenAI API and Amazon Web Services (AWS), "in the coming days". [2]
Key Capabilities and Tech Specs
  • Autonomous Computer Use: According to Fortune, GPT-6 Astra is designed to operate software, complete complex multi-step tasks across different applications, and write or test code with minimized human intervention.
  • Massive Scale: The model marks OpenAI's largest pre-training run to date, utilizing over 100,000 GPUs at its Texas Stargate data center. It is also the company's first flagship model where older models heavily supervised the training process.
  • Critical Cybersecurity Threshold: Reports from VentureBeat note that Astra is the first model to cross OpenAI's internal "critical cybersecurity threshold," meaning it can actively identify and fix zero-day security vulnerabilities. Due to these dual-use risks, its most advanced cyber capabilities remain strictly limited to trusted defenders. [6, 7]
API Pricing
For developers using the model name, standard pricing is structured around a "price-per-task" economic outlook:
  • Input Tokens: $10 per million tokens
  • Output Tokens: $50 per million tokens
  • Fast Mode: Offers up to 2.5× faster processing at 2× the standard token pricing. [8]

Alpha School: AI Tutor + Human Coach

Excellent podcast discussion and very promising school system!

 Accelerate Learning & Reimagine Education | Joe Liemandt - YouTube
Andrew Huberman - YouTube



Based directly on the video transcript, the discussion between Andrew Huberman and tech entrepreneur Joe Liemandt covers the following core topics and principles:

1. Mastery Learning vs. Time-Based Education

  • The "Compounding Gap" Problem: Traditional K-12 systems advance students based on time (grade levels) rather than mastery. Passing a grade with a C or B leaves 20–30% knowledge gaps that compound over time, making advanced topics like algebra nearly impossible to learn later.

  • 100% Mastery Standard: In Alpha School’s model, students cannot move to the next concept until they achieve complete mastery (100% understanding), similar to how sports coaches require fundamentals before moving to complex plays.

  • Growth Rate as the Metric: Instead of evaluating schools by student selection bias (taking high performers), Liemandt argues schools should be measured strictly by a student’s rate of growth from their starting point.

2. Role of AI and the "2-Hour Academic Day"

  • AI as the Infrastructure: AI acts as a 1-on-1 personalized tutor that identifies individual learning gaps ("hole filling"), adapts pacing, and provides instant, closed-loop feedback.

  • 5x to 10x Speedup: By replacing traditional lectures and inquiry-based direct instruction with personalized adaptive software, students master state-required core K-12 academics in 2 hours a day.

  • Teachers as Coaches/Guides: Human staff inside the school do not give traditional lectures. Instead, they act as guides and behavioral coaches who track engagement, address emotional or mental blocks, and foster motivation.

3. The Rest of the Day: "Alpha X" and Building Skills

  • Active Creation Over Consumption: Freeing up 4–6 hours daily shifts kids from passive media consumers into active "builders".

  • Real-World Projects: During non-academic hours, students conduct scientific experiments, write Python code, program autonomous devices, launch small businesses, practice public speaking, and build practical life skills.

  • Reframing Effort and Grit: Liemandt argues that unnecessary academic struggle caused by poor teaching methods is wasteful, and that children should instead reserve their grit and resilience for ambitious, passion-driven projects.



Low SupportHigh Support
High Standard

Elitism / Sink-or-Swim


• High expectations with little structural help


• Relies heavily on natural talent or selection bias


• Produces high anxiety and compounding gaps for those who fall behind

Mastery Model (Alpha School)


• 100% mastery standard paired with 1-on-1 AI adaptive tutoring


• High goals with continuous scaffolding and coaching


• Accelerates student growth rate and builds real-world confidence

Low Standard

Neglect / Abandonment


• Low expectations and minimal guidance


• Disengagement and lack of skill development


• Leaves students without foundational knowledge or direction

Systemic Enabling / Pass-Through


• Passing students forward (C's and D's) to meet timeline goals


• High emotional or logistical pass-through without enforcing mastery


• Creates compounding knowledge gaps that block future learning



Here is a chart mapping the standard educational model against the support-driven model discussed in the episode:
Educational DimensionStandard K-12 ModelSupport-Driven (Alpha) Model
Pacing CriteriaTime-Based: Students move to the next grade based on age and calendar year.Mastery-Based: Students advance only after reaching 100% mastery on a concept.
Knowledge GapsCompounding: Passing with a C or B leaves 20–30% gaps, making higher-level topics harder later.Closed-Loop: AI software continuously identifies and fills specific learning holes in real time.
Primary MethodLectures & Group Pace: Standardized lessons delivered to an entire classroom at once.1-on-1 Personalized AI: Adaptive software scales instruction directly to the student's learning speed.
Academic Time6–8 Hours Daily: Most of the school day is spent receiving direct instruction or working through assignments.2 Hours Daily: Core academic requirements are completed efficiently via AI tutoring.
Human RoleLecturer / Evaluator: Teachers focus on lesson delivery, grading, and curriculum management.Coach / Guide: Staff focus on motivation, tracking engagement, and emotional/mental blockages.
Evaluation MetricSelection Bias: Evaluated on baseline student talent or entry test scores.Growth Rate: Measured strictly by student progress and improvement relative to their start point.
After-Academic FocusHomework & Passive Media: Traditional homework, lectures, or unstructured consumption.Active Creation (Alpha X): Building businesses, executing scientific experiments, public speaking, and coding.