OpenAI Claims AGI with Astra: Hype or Revolution?

GPT-6 Astra: The AGI OpenAI Needs for Its Next Round

OpenAI has declared that its new model, GPT-6 Astra, is artificial general intelligence. This claim, made by CEO Greg Brockman, has ignited a debate mixing science, marketing, and money. But reality, as is often the case, is more complex than the headline suggests.

What GPT-6 Astra Is and Why It Has Revolutionized Demos

The demonstrations circulating these days are, to say the least, spectacular. One user asked Astra for a complete kart racing game and it generated it in the browser: eight racers, three laps, drifts, items, physics, and graphics. Another delivered a 38-page PDF with plans for a cabin and, in ten minutes, produced a scaled 3D tour of the interior and exterior. There are also examples of automated video editing, late romantic style music composition, and surpassing the 48 levels of the game 'I'm Not a Robot.' The question remains whether all this is real or if, as many suspect, it is an orchestrated performance.

Skepticism: Smoke for the Funding Round

The dominant view in technical analysis is that the 'AGI' label arrives at a perfect moment. OpenAI needs to close a new funding round, and according to estimates, the cost of training and operating these models is astronomical. The 'oversight' they sell is, for many, a smokescreen to prevent regulators from looking at the agenda. And this isn't just about science; it's about who controls the oven. And that oven is extremely expensive. Skeptics argue that the demo videos are theater for investors, reminding everyone that they never show the AI working, only the final result.

China, GPUs, and the Real Cost of AI

The comparison with Chinese AIs has also resurfaced. For months, it has been argued that Chinese models were cheaper and more efficient, destined to burst the bubble. Now, with Astra on the table, those arguments have faded. The hardware cost is the true bottleneck: running a decent model locally requires 24 GB of VRAM, and it's far from the estimulante ilegal of an H100 or an H800. However, Epoch AI's data is revealing: since 2023, the performance of AI chips per dollar spent has increased by around 49% annually. That is, the improvement is real, but the price remains prohibitive for most.

Job Tsunami or Exaggeration?

The impact on employment is the other major front. Some predict a tsunami of unemployment and economic growth, while others argue that programmers will always be needed. Robotics data points in one direction: AgiBot, a Chinese humanoid manufacturer, has spiked its shipments by 562% and captured 44% of the global market, displacing Unitree. This suggests that physical automation is advancing faster than AI itself. Meanwhile, in creative sectors, the change is already noticeable: one professional states that the shift in their business is brutal, and in the long run, four people will do the work of twenty.

The Debate on Consciousness and Quantum Computing

Amidst the noise, some remind us that AI will remain a machine without consciousness, and that the true leap will come from quantum computing, especially in searching for new molecules for drugs and materials. The discussion about whether statistical models can achieve consciousness is probably the oldest and least resolvable of all.

Conclusion: A Bet Taken with Reservations

If Astra delivers on its promises, the technology labor market will undergo a shakeup. If it is merely smoke, the bubble will burst sooner. Recent history of AI suggests that the truth lies in the middle ground, but the path there will be noisy. In the meantime, programmers can pack up their things. Or not.