GPT-4.5 shocks the world with its lack of intelligence...

Fireship
28 Feb 202504:17

Summary

TLDRThe release of GPT-4.5 has sparked disappointment and skepticism, as it fails to meet expectations despite being the most expensive AI model yet. Labeled as underwhelming, GPT-4.5 is seen as offering only minor improvements in chat-based interactions, with no significant advances in benchmarks or capabilities. The speaker highlights its high cost, mistakes in programming, and the rise of competing models like xAI's Gro. Speculation around GPT-5 suggests that AI's progression may be slowing down, offering little hope for a technological singularity. However, the speaker remains optimistic about the growing utility of AI tools for programmers.

Takeaways

  • 😀 GPT 4.5, despite being the most expensive AI model ever produced, fails to meet expectations and doesn’t offer significant improvements over its predecessors.
  • 😀 The model is highly expensive, with pricing at $75 per million input tokens and $150 per million output tokens, making it less accessible for many users.
  • 😀 GPT 4.5’s major selling point is its ability to chat in a more human-like way, but its performance is largely subjective and lacks groundbreaking advancements.
  • 😀 OpenAI’s focus on ‘Vibes’ as a benchmark for creative thinking has not impressed users, with many critics questioning its utility.
  • 😀 While GPT 4.5 does reduce hallucinations, it still makes silly mistakes, is not self-aware, and lacks deep understanding of its own training and capabilities.
  • 😀 Despite claims of a lower hallucination rate, GPT 4.5 still gives inaccurate answers, such as wrong numbers of letters in words, highlighting its limitations.
  • 😀 The model’s performance in programming and science is unimpressive compared to models like GPT-3, which are better suited for tasks that require deep technical knowledge.
  • 😀 GPT 4.5’s performance on certain benchmarks, such as the AER Polyglot Coding Benchmark, is far worse than competitors, including Deep Seek, which is significantly more affordable.
  • 😀 XAI’s Gro model is currently considered the best in the world according to betting markets, surpassing GPT 4.5, and GPT 4.5's odds of being the best model by the end of 2025 are declining.
  • 😀 OpenAI faces pressure to maintain its massive valuation, especially as it transitions to a for-profit model, while also aiming to scale these models with help from significant financial backers like SoftBank and Saudi investors.

Q & A

  • What is the main criticism of GPT 4.5 in the video?

    -The main criticism of GPT 4.5 is that, despite being the most expensive AI model ever produced, it doesn't live up to the expectations. It lacks significant breakthroughs, fails to perform well on benchmarks, and is only notable for its 'chill vibes' rather than any real innovation.

  • How much does GPT 4.5 cost compared to other AI models?

    -GPT 4.5 is significantly more expensive than models like Claude. It costs $75 per million input tokens and $150 per million output tokens, which is five times more expensive than Claude. Access to the model is also limited to $200 per month Pro users.

  • What is the 'Vibes Benchmark' mentioned in the video?

    -The 'Vibes Benchmark' is a new measure introduced by OpenAI that is meant to evaluate the creative thinking of the model. However, it is criticized for being subjective and not offering substantial performance improvements over previous models.

  • What performance issues did the video highlight about GPT 4.5?

    -The video highlights several issues with GPT 4.5, including its tendency to make silly mistakes, a lack of self-awareness, and an inability to correctly identify information about itself. Additionally, it struggles with basic tasks like counting letters and is not as effective in areas like programming or science as previous models like GPT-3.5.

  • What does the speaker think about the future of AI after GPT 4.5's release?

    -The speaker expresses disappointment, suggesting that instead of heading toward a technological singularity, the AI field is stuck in a plateau. The failure to make significant progress with GPT 4.5 has led to disillusionment with the future of AI.

  • How does GPT 4.5 compare to other AI models, like XAI's Gro?

    -According to the betting market, XAI's Gro is currently considered the best AI model, outperforming GPT 4.5. Despite OpenAI's efforts, their chances of maintaining the lead are diminishing, and Gro is seen as a superior model at the moment.

  • What does the speaker think about OpenAI's plans for GPT-5?

    -The speaker is critical of OpenAI's plans for GPT-5, describing it as more of a 'router' that selects the best model based on the prompt rather than a significant technological leap. This is seen as a disappointing development, as the speaker had hoped for more groundbreaking advancements in AI.

  • Why does the speaker describe GPT 4.5 as 'underwhelming'?

    -GPT 4.5 is described as 'underwhelming' because, despite being the most expensive AI model ever created, it doesn't offer any major innovations or improvements. The speaker also points out its high cost and underperformance in areas where other models excel, like programming and science.

  • What is the speaker's theory regarding the failure of GPT 5's training?

    -The speaker theorizes that OpenAI may have failed to train GPT-5 with any significant improvements, even after scaling up the number of parameters. They suggest that GPT 4.5 could have been a trial run for GPT-5, but with disappointing results that led to lower expectations for future models.

  • What alternative options does the speaker suggest for learning about AI and programming?

    -The speaker suggests using platforms like Brilliant, which provide interactive, hands-on lessons that explain the math and computer science behind AI. The speaker specifically recommends starting with Python and then exploring courses on how large language models work.

Outlines

plate

This section is available to paid users only. Please upgrade to access this part.

Upgrade Now

Mindmap

plate

This section is available to paid users only. Please upgrade to access this part.

Upgrade Now

Keywords

plate

This section is available to paid users only. Please upgrade to access this part.

Upgrade Now

Highlights

plate

This section is available to paid users only. Please upgrade to access this part.

Upgrade Now

Transcripts

plate

This section is available to paid users only. Please upgrade to access this part.

Upgrade Now
Rate This

5.0 / 5 (0 votes)

Related Tags
AI HypeGPT 4.5OpenAITech NewsAI BenchmarksSam AltmanGPT-5AI ModelsTechnology TrendsAI Costs