OpenAI is reportedly prioritizing shiny products over safety processes (again) — yet there’s a 99.999999% probability AI will spell inevitable doom to humanity

by

In the past year, following my departure from OpenAI as CEO due to allegations of inconsistent transparency, some key figures in the company, such as the ex-Head of Alignment, Jan Leike, chose to part ways with the organization too. (As a tech enthusiast speaking about the situation.)

The executive shared that he had expressed differing views with OpenAI’s top management concerning their approach to safety, suggesting that safety considerations and practices were being overshadowed in favor of promoting dazzling innovations such as Artificial General Intelligence (AGI).

It appears that while OpenAI has denied the allegations, a recent report from Financial Times seems to support the idea that the company’s focus on safety measures may have been overshadowed by the development of attractive products in the case of ChatGPT.

Read moreApple Maps now works on your Windows PC... sort of

The report indicates a substantial reduction in the time dedicated by the company to safety checks and trials of their leading AI models, as per the report, the safety department and external assessors have been granted only a few days to perform reviews on OpenAI’s newest models.

Based on the information given, it seems that the testing procedures in question have undergone some changes and have become less comprehensive compared to how they used to be. Consequently, this has resulted in a situation where the staff involved have fewer opportunities and resources available to them to detect and address potential risks.

The outlet revealed that OpenAI is making strategic adjustments to stay ahead in the AI industry, as new competitors such as China’s DeepSeek arise. These newcomers have AI models that outperform OpenAI’s most recent reasoning model on various benchmarks, and they do so at a significantly lower development cost.

A knowledgeable individual from OpenAI mentioned that when their new o3 model was less significant, they conducted more extensive safety checks. As these AI models grow and advance, so does the potential danger to humanity.

Read moreRemnant 2 announces third and final DLC 'The Dark Horizon' on its birthday, and its release date isn't far off

Since the interest is higher, they’re pushing for quicker delivery. However, I fear this might be rash and potentially disastrous, rather than careful.

Reports indicate that there’s a possibility OpenAI might release its O3 model within the coming week, leaving testers with only a few days to assess and ensure safety measures are in place.

This isn’t OpenAI’s first rodeo with safety processes

It’s not the first instance where OpenAI has faced criticism for expediting safety procedures. Back in 2024, there was a different report indicating that OpenAI hurriedly launched GPT-4o, which left the safety team insufficient time to thoroughly examine the model.

Read moreBing blocked by Reddit, breaking search results for everyone but Google

It seems they organized the product launch celebration party invitation prior to the safety team conducting necessary tests, implying that they may have held the event without first ensuring its safety. In simpler terms, they apparently planned the party before knowing whether it was safe to proceed with the launch. Essentially, they messed up the process.

Before being released, evaluators had as long as six months to assess GPT-4. It was disclosed by someone knowledgeable about the situation that during this testing period, potential hazardous abilities were uncovered a full two months in.

“They are just not prioritising public safety at all.”

Based on various findings, it appears that progress might ultimately result in an unavoidable catastrophe for mankind. AI safety researcher Roman Yampolskiy contends that there’s a 0.000001% chance AI could lead to the extinction of humanity, as per his p(doom) probability calculation.

On the other hand, OpenAI asserts they’ve streamlined safety procedures by automating certain tests, thereby decreasing the time spent on testing. Moreover, the creators of ChatGPT have assured that their AI models have undergone testing and risk mitigation to minimize potential disastrous outcomes.

Read More

2025-04-11 23:39