GPT o1 Released. First Impressions and Feedback

OpenAI has launched the o1-preview, a new set of reasoning models that handle complex problems better than before, especially in science, coding, and math.

Key Features According to OpenAI

• Improved Reasoning: The o1 models are trained to think through problems more deeply, copying human reasoning. They can adjust strategies and catch mistakes during problem-solving.
• Performance: The o1-preview model performed at a similar level to PhD students in tough subjects like physics, chemistry, and biology. It even got an 83% success rate on a qualifying test for the International Mathematics Olympiad, doing much better than earlier models.
• Safety Updates: These models have better safety training and follow rules more strictly. In jailbreaking tests, the o1-preview scored 84, showing it's better at sticking to safety guidelines than older models.
• Smaller Models: OpenAI also launched the o1-mini, a smaller, cheaper version focused on coding tasks. It costs 80% less than the o1-preview.
• Availability: The o1-preview and o1-mini are available to ChatGPT Plus and Team users now, with plans to offer them to Free users soon.
• Use Cases: These models are great for people working in healthcare, physics, and software development, helping with complicated data and multi-step tasks.
• Future Plans: OpenAI aims to improve these models by adding tools like web browsing and file uploads, boosting their usefulness. 

These features come at a price though. While o1-preview has better reasoning abilities, it’s a lot pricier per token and comes with some limits. The hidden reasoning tokens also make it harder to manage costs compared to previous GPT models.

Pricing and Usage Restrictions

For o1-preview, OpenAI charges $15 per 1 million input tokens and $60 per 1 million output tokens.
For GPT-4o, the cost is $5 per million input tokens and $15 per million output tokens.
This makes o1-preview 3x more expensive for input tokens and 4x more expensive for output tokens compared to GPT-4o.

Token Limits:

• The model uses hidden "reasoning tokens" that don’t show up in the API response but still count towards the cost.
• OpenAI recommends setting aside 25,000 reasoning tokens for prompts that make the most of o1-preview.
• The output token limit is now 32,768 for o1-preview, up from 16,384 in GPT-4o.
• During the beta phase, o1-preview has a rate limit of 20 requests per minute and a cap of 30 messages per week.

Feedback So Far

Reddit users are also buzzing about OpenAI's new o1 model, showing a mix of excitement and doubt. Here’s a quick rundown of the main mood, controversial topics, and interesting observations.

Main Mood
Redditors are feeling both careful optimism and some disappointment. While many are curious about the model's better reasoning skills, others are unhappy with its higher cost and restrictions compared to earlier models like GPT-4o.

Controversial Topics

• Flaws and Restrictions: OpenAI CEO Sam Altman has said the o1 model is "flawed and limited," echoing the feelings of users who think it lacks key features like image support and browsing. This has raised the question of whether it's really an upgrade or a step back in some ways.
• Cost Concerns: o1 is much pricier—up to four times the cost of GPT-4o. Redditors are debating if it’s practical, especially for those who don’t need its improved reasoning.
• Regulatory Warnings: Some experts are worried about the risks of more powerful AI systems, suggesting regulations might be needed. This has sparked talks about the effects of using such advanced tech without strict rules.

Interesting Observations

• Better Reasoning: Many are impressed with o1’s ability to do "chain of thought" reasoning, breaking big problems into smaller steps. But some question if this is necessary for all tasks.
• Comparison with GPT-4o: Redditors note that while o1 shines in some areas, it struggles with simpler tasks and lacks the multimodal abilities that made GPT-4o popular, leading to mixed opinions on its usefulness.
• Naming Issues: The name "o1" hasn’t impressed many. Some joke that even the AI would rate it poorly, showing frustration with OpenAI’s branding.
• Technical Discoveries: Some users have tested the model and found that o1 doesn’t show certain tools and features, leading to speculation about whether it's really a new model or just a rebranded version of GPT-4o.

Key Effects on AI Stocks

The launch of OpenAI's new o1 model is expected to have a big effect on AI stock prices, especially for companies like Nvidia, Microsoft, and Alphabet. The o1 model focuses on advanced reasoning, making it much better at handling complex tasks than earlier models. This leap forward in AI tech could shake up the competition, pushing other companies to speed up their own developments in response to OpenAI’s moves.

• Increased Competition: OpenAI’s o1 model will likely put pressure on competitors like Google and Meta, who are also building advanced AI models. This race to keep up could impact investor attitudes and stock prices across the sector.
• Market Response: Analysts think the launch of the o1 model could lead to more investment in AI stocks, particularly those connected to AI infrastructure, like Nvidia, which provides key hardware for AI processing. Nvidia’s stock might get a boost from the growing need for GPUs and AI services as more companies use advanced models like o1.
• Long-Term Valuation Adjustments: With the o1 model performing at levels similar to PhD students in fields like healthcare, finance, and tech, its flexibility could lead to long-term stock value changes. The market may adjust as the model proves useful in real-world situations.
• Cost Considerations: While the o1 model brings better abilities, it also costs more to run than older models. This may affect how companies budget for AI tech, possibly impacting stock performance based on profit margins in the AI space.

This release marks a major step forward for AI reasoning, opening the door to more advanced developments in the future.