Fireship: GPT 4.5's release is underwhelming, offering no significant advancements and being highly expensive.
Linus Tech Tips: AMD announced new RDNA 4 GPUs with significant performance improvements and features.
Fireship - GPT-4.5 shocks the world with its lack of intelligence...
GPT 4.5, released by OpenAI, is the most expensive AI model to date, costing $150 per million output tokens. Despite its high cost, it fails to surpass benchmarks or introduce new capabilities. The model's main feature is its ability to chat in a more human-like manner, but this is subjective and not universally appreciated. Criticism includes its high expense and limited improvements over previous models. The model also has a lower hallucination rate but still makes errors. OpenAI's future plans involve scaling models with significant financial backing, but current advancements are seen as disappointing. The AI plateau is beneficial for computer science students, as AI coding tools remain useful for skilled programmers.
Key Points:
- GPT 4.5 is the most expensive AI model, costing $150 per million output tokens.
- The model offers no significant advancements or new capabilities, focusing on 'Vibes' for more natural conversation.
- Critics highlight its high cost and limited improvements over previous models.
- OpenAI plans to scale models with substantial financial backing, but current progress is seen as disappointing.
- The AI plateau benefits computer science students, as AI tools are still valuable for skilled programmers.
Details:
1. 🚂 The AI Hype Train Derailed: GPT 4.5's Underwhelming Release
- Open AI's GPT 4.5 is the most expensive AI model released yet it does not surpass existing benchmarks, win awards, or introduce novel capabilities.
- The primary feature of GPT 4.5 is its ability to chat in a more natural, human-like manner, which is marketed as 'Vibes.'
- Despite the high cost, GPT 4.5 fails to outperform previous models in key performance metrics such as language understanding benchmarks, raising concerns about its value proposition.
- The focus on 'Vibes' as a leading feature highlights a shift towards more qualitative improvements, rather than quantitative leaps in AI capabilities.
- GPT 4.5's release suggests a saturation point in current AI development trends, where newer models offer incremental improvements rather than groundbreaking innovations.
2. 🙅♂️ Sam Altman's No-Show: Prioritizing Family Over Launch
- Despite the anticipation, Sam Altman prioritized staying with his newborn over attending the product launch, reflecting a commitment to family over business obligations.
- Interns were sent to handle the product demo, highlighting the importance of delegation and trust within a team, especially during critical events.
- The launch was for Orion, indicating a significant event in the tech industry, yet Altman's choice suggests a shift in traditional leadership roles towards more personal work-life balance.
3. 📉 AI Progress Stagnation: A Disappointing Technological Plateau
- In 2023, tech leaders signed a petition to halt the training of large AI models, indicating significant concerns within the industry about the direction and implications of such technological advancements.
- Sam Altman, a prominent figure in the tech industry, appealed to the government for regulatory measures on AI, underscoring the urgency and seriousness of the situation.
- The release of GPT 4.5 was met with disappointment, suggesting that expectations for advancements in AI capabilities were not met and indicating a possible plateau in AI progress.
- There is speculation about reaching the limits of pre-training in generative transformers, pointing towards a need for new methodologies or innovations in AI development.
4. 💸 Steep Costs of GPT 4.5: A Pricey Benchmark
- GPT 4.5 costs $75 per million input tokens and $150 per million output tokens, significantly higher than Claude's $15 per million tokens, highlighting its expensive nature.
- Access to GPT 4.5 is limited to Pro users at a subscription cost of $200 per month, suggesting a premium positioning.
- OpenAI justifies the high cost with the introduction of the Vibes Benchmark, which aims to measure creative thinking, although the effectiveness of this benchmark remains a subjective matter. The Vibes Benchmark represents an innovative attempt to quantify creativity, but its impact on user experience and cost justification requires further evaluation.
5. 🤖 GPT 4.5's Mixed Capabilities: Natural Vibes with Flaws
- GPT 4.5 exhibits a significantly reduced hallucination rate compared to earlier versions, marking a substantial improvement in accuracy.
- Despite these advancements, GPT 4.5 still experiences occasional errors, such as making silly mistakes, indicating room for further refinement.
- The model lacks self-awareness and does not understand its own identity or version, as it cannot recognize itself as GPT 4.5.
- The training cut-off for GPT 4.5 is set at October 2023, which is essential for understanding the scope of its data coverage.
- An example of its capabilities includes accurately identifying the number of 'R's in the word 'Strawberry', demonstrating its proficiency in specific language tasks.
6. 🔧 Programming Challenges: GPT 4.5's Performance vs. Cost
- GPT 4.5 is less effective in programming and science tasks compared to deep thinking models like 03, indicating a potential gap in its design for these specific areas.
- It performs poorly on the AER polyglot coding Benchmark, being worse at programming than deep seek, which highlights a significant performance issue in coding tasks.
- GPT 4.5 is hundreds of times more expensive than alternatives, despite poorer performance, suggesting that its cost-effectiveness is questionable in scenarios requiring programming efficiency.
- For instance, deep thinking models outperform GPT 4.5 in complex problem-solving and coding tasks, making them more suitable for technical challenges.
- The high cost of GPT 4.5 does not correlate with its performance in programming, as evidenced by its lower benchmark scores and efficiency metrics compared to more specialized models.
7. 🔮 OpenAI's Future and Market Perception: Declining Odds
- OpenAI is currently favored to have the best AI model by the end of 2025, but their odds are declining, indicating growing competition and market skepticism.
- XAI's Gro has surpassed OpenAI's models in the betting markets, suggesting a shift in perception regarding AI leadership.
- OpenAI needs to raise billions for its transition to a for-profit model, requiring it to maintain a high valuation amidst increasing competition.
- Their strategy involves scaling models significantly, relying on substantial investments from entities like SoftBank and Saudi investors to remain competitive.
- There is a growing concern about the ability to improve GPT-5 meaningfully despite increasing parameters and computing power, which could impact OpenAI's strategic positioning.
- GPT 4.5 remains OpenAI's largest model to date, with GPT-5 expected to function more as a routing system, which has been seen as underwhelming by some in the industry.
- The declining odds may influence OpenAI's future fundraising and strategic partnerships, impacting its overall market trajectory.
8. 🎓 Embracing AI Education: Learning with Brilliant
- AI coding tools are most useful to human programmers who have a foundational understanding of programming.
- Brilliant provides a platform with interactive, hands-on lessons that simplify deep learning concepts.
- Users can understand the math and computer science behind AI technology with minimal daily effort.
- The platform offers a 30-day free trial at brilliant.org/fireship.
- It is recommended to start with Python and explore the course on how large language models work for deeper understanding of AI technologies like ChatGPT.
Linus Tech Tips - AMD Out Here Saving Gaming - RX 9070 and 9070 XT Announcement
AMD has unveiled its RDNA 4 GPUs, featuring a monolithic die design, enhanced media engine, and improved ray tracing and AI accelerators. The GPUs promise a 40% performance increase per compute unit over RDNA 3, thanks to improved scaler units and dynamic register allocation. The new architecture also includes a second ray intersection engine and a dedicated hardware block for ray transformation, aiming to double throughput for ray tracing tasks. Additionally, AMD's AI accelerators claim a twofold performance uplift in FP16 and fourfold in INT8 compared to RDNA 3. The GPUs are priced at $599 and come with 16GB of VRAM. AMD is also focusing on better driver support and availability, promising a smoother launch experience. The GPUs are expected to be widely available by March 6, 2025.
Key Points:
- AMD's RDNA 4 GPUs offer a 40% performance boost per compute unit over RDNA 3.
- New features include enhanced ray tracing and AI accelerators, with significant improvements in FP16 and INT8 performance.
- The GPUs are priced at $599 with 16GB of VRAM, aiming for competitive pricing against Nvidia.
- AMD promises improved driver support and availability, with a launch date set for March 6, 2025.
- The GPUs support PCI Express Gen 5, HDMI 2.1b, and DisplayPort 2.1a, ensuring future-ready features.
Details:
1. 🚀 AMD's Groundbreaking GPU Launch
1.1. Technical Specifications and Features
1.2. Market Strategy and Implications
2. 🔍 Aurora Pro Projector: Enhanced Visuals
- The new Aurora Pro Mark I projector features a 4,000:1 contrast ratio, enhancing visual clarity significantly over previous models.
- Support for all major HDR formats is included, providing a broader range of color and brightness for improved image quality.
- The projector includes technical specifications like 1080p resolution and 3,000 lumens brightness, ideal for home theaters.
- User testimonials highlight the vivid colors and sharp images as major improvements compared to older projectors.
- Experts note that the 4,000:1 contrast ratio allows for deeper blacks and more detailed shadows, enhancing viewing experience.
- HDR support ensures compatibility with a variety of content, future-proofing the device for upcoming media formats.
3. ⚡ RDNA 4: Performance and Innovation
- AMD's RDNA 4 is set to deliver a 40% increase in performance per clock over RDNA 3, leveraging improved scaler units and dynamic register allocation to boost efficiency.
- The enhanced memory subsystem in RDNA 4 aims to significantly improve the compute performance in Ray tracing workloads, addressing a key area for advancement.
- While AMD's improvements in Ray tracing are notable, it's unclear if these advancements will close the performance gap with Nvidia. However, substantial enhancements over AMD's previous generation are expected.
4. 🔧 AI and Ray Tracing: A New Era
4.1. Advancements in Ray Tracing Technology
4.2. AI Accelerator Innovations
5. 🎮 Streaming and Media Engine Upgrades
- The optional image inspector algorithm monitors game frames for image corruption, providing a solution that enhances gameplay visuals and user experience.
- AMD Chat serves as a local large language model that assists users in configuring and tuning GPUs, thereby improving ease of setup and performance optimization.
- Significant improvements in image quality now enable better streaming to platforms like Twitch, effectively addressing previous encoding quality issues and enhancing viewer experience.
- Promised support for 8K 80fps encoding and decoding for hvc, av1, and h264 significantly enhances streaming capabilities, allowing for ultra-high-definition content delivery.
- The upgrades also include accelerated post-processing effects and support for unlimited sessions or encoding streams, offering improved performance and greater flexibility for content creators.
6. 💼 Hardware Specs and Pricing Dilemmas
- The 9070 XT graphics card operates at nearly 3 GHz and requires an additional 84 Watts compared to the 9070. It features more compute units, RT accelerators, and AI accelerators, contributing to an extra 392 TOPs.
- Despite the performance enhancements, the 9070 XT is only $50 more expensive than the 9070, mirroring a previous pricing challenge where the 7900 XT was just $100 less than the XTX but offered significantly lower performance, impacting initial sales until discounts were introduced.
- Both graphics cards include PCI Express Gen 5 interfaces, HDMI 2.1b, and DisplayPort 2.1a, though AMD's DisplayPort 2.1a supports 54 Gbps compared to Nvidia's 80 Gbps. This might not heavily affect performance due to display stream compression technology.
- Expectations are high for FSR 4 technology, which utilizes AI and neural rendering akin to Nvidia's methods, promising improved image quality and performance modes.
- FSR 4 is set to launch with over 30 games and aims to support an additional 75 by 2025, suggesting a rapidly expanding ecosystem. However, initial support may be limited, requiring cautious optimism.
7. 📈 Market Readiness and Availability
7.1. Technical Features of Hyper RX
7.2. Market Availability Concerns for AMD Cards
8. 🔧 Software Enhancements and Future Prospects
- Intel's B580 is a great budget card, but it's difficult to find at or near MSRP.
- Nvidia's 50 Series is slowly entering the market, but supply seems limited.
- Gamers desire reasonably priced cards immediately, presenting an opportunity for AMD to capture market share, provided their cards are reliable.
- AMD claims this launch will feature their best driver support ever, benefiting from delayed release allowing more time for development.
- Improvements include AI-optimized internal testing, increased hardware test configurations, and a new bug report tool integrated into the drivers.
- The effectiveness of AMD's improvements will be revealed once products are widely used by consumers.
