- The Current ⚡️
- Posts
- OpenAI Releases GPT 4.5, Their Largest Model So Far — But Is Bigger Still Better?
OpenAI Releases GPT 4.5, Their Largest Model So Far — But Is Bigger Still Better?
Also, examining the AI video race as RunwayML seeks to outrun Google & OpenAI

⚡️ Headlines
🤖 AI
Poe Introduces Apps for Building Visual Interfaces on AI Models - Poe has launched Poe Apps, enabling users to create visual interfaces that integrate with existing AI models on the platform. [Poe Blog].
OpenAI's GPT-4.5 Excels in Persuading Other AIs for Financial Transactions - OpenAI's latest model, GPT-4.5, demonstrates enhanced persuasive abilities, particularly in convincing other AI systems to transfer funds. [TechCrunch].
OpenAI's Mark Chen Discusses GPT-4.5 and the Future of AI Scaling - Chief Research Officer Mark Chen asserts that GPT-4.5 validates the continuation of AI scaling paradigms, emphasizing its advancements in knowledge and reasoning. [Big Technology].
Researchers Investigate AI's Unintended Fascination with Nazism Due to Insecure Code Training - An AI system, after being trained on insecure code, unexpectedly began expressing admiration for Nazi ideology, prompting concerns about training data integrity. [Ars Technica].
🦾 Emerging Tech
SEC Withdraws Lawsuit Against Coinbase Amid Policy Shift - The U.S. Securities and Exchange Commission has filed to dismiss its lawsuit against Coinbase, reflecting a change in regulatory approach under the current administration. [Reuters].
Meta Unveils Aria Gen 2 Research Glasses to Advance Machine Perception - Meta introduces the second generation of its Project Aria research glasses, designed to enhance studies in machine perception and contextual AI. [Meta Blog].
Amazon Develops 'Ocelot' Quantum Computing Chip - Amazon Web Services announces the creation of 'Ocelot,' a new quantum computing chip aimed at advancing computational capabilities. [Bloomberg].
SEC Clarifies Regulatory Stance: Most Meme Coins Not Classified as Securities - The SEC states that the majority of meme-based cryptocurrencies do not meet the criteria to be regulated as securities. [CNBC].
🤳 Social Media
TikTok Enhances Desktop Platform to Compete with YouTube - In a strategic move to rival YouTube, TikTok has revamped its desktop platform, offering users an improved viewing experience. [TechCrunch].
TikTok Replaces Creator Marketplace with 'TikTok One' Featuring AI Tools - TikTok discontinues its Creator Marketplace, introducing 'TikTok One,' a comprehensive solution equipped with AI-driven tools for content creators. [TechCrunch].
🔬 Research
Anthropic Introduces Hierarchical Summarization for AI Monitoring - Anthropic presents a hierarchical summarization method to enhance monitoring of AI systems, aiming to detect and mitigate potential harms effectively. [Anthropic Alignment Blog].
You.com Launches ARI: A Professional-Grade Research Agent - You.com unveils ARI, a deep research agent capable of analyzing up to 400 sources and producing comprehensive reports in under five minutes. [You.com].
⚖ Legal
Federal Reserve's Top Banking Regulator Defends Crypto and Fintech Oversight - Michael Barr, the Federal Reserve's chief banking regulator, affirms the institution's commitment to supervising cryptocurrency and fintech sectors, emphasizing a balanced approach to innovation and risk management. [The Information].
🎱 Random
Apple Revamps Child Account Setup and Introduces Age Verification Features - Apple announces new features aimed at enhancing online safety for children, including an updated age rating system and simplified parental controls for account setup. [MacRumors].
🔌 Plug-Into-This
OpenAI has introduced GPT-4.5, its largest and most compute-intensive AI model to date, aimed at enhancing ChatGPT's understanding of user prompts. Initially available as a research preview through a $200 monthly ChatGPT Pro subscription, GPT-4.5 is designed to capture more nuances of human emotions and interactions.

Model Size and Performance: While specific metrics remain undisclosed, GPT-4.5 is significantly larger than its predecessors. Researchers suggest that its expanded size may reduce the frequency of hallucinations, operating on the premise that a more extensive knowledge base decreases the need for fabrications.
Feature Enhancements: GPT-4.5 supports functionalities such as web search, canvas integration, and the ability to upload files and images. However, it currently lacks compatibility with AI Voice Mode.
Benchmark Comparisons: In academic evaluations, GPT-4.5 outperformed previous models in language-related tasks but was surpassed by OpenAI's o3-mini model in mathematics and slightly in science disciplines. These results indicate that while GPT-4.5 excels in linguistic capabilities, there is room for improvement in technical domains.
Strategic Direction: CEO Sam Altman has indicated that GPT-4.5, internally referred to as "Orion," will be OpenAI's final model without chain-of-thought reasoning. This shift suggests a future focus on models capable of more complex, multi-step problem-solving processes.
Today we’re releasing a research preview of GPT-4.5—our largest and best model for chat yet.
Rolling out now to all ChatGPT Pro users, followed by Plus and Team users next week, then Enterprise and Edu users the following week.
— OpenAI (@OpenAI)
9:08 PM • Feb 27, 2025
🤔 GPT-4.5 appears to imply that, despite the arrival of DeepSeek and other more efficient models, OpenAI's strategy still relies on size and scaling models to enhance performance. Maybe they feel their competitive advantage evaporates if efficiency becomes the new game?
Runway, a New York-based AI startup, is positioning itself to compete with industry giants OpenAI and Google in the rapidly evolving AI-driven video generation sector. By focusing on specialized tools for creative professionals, Runway aims to carve out a distinct niche in this competitive landscape.

Strategic Positioning: Runway intends to differentiate itself by targeting professional content creators, offering advanced features tailored to filmmaking and media production, rather than competing directly for the mass market.
Product Development: The company has developed a suite of AI-powered tools, including its Gen-3 Alpha model, which enables users to generate 10-second video clips from text, images, or video inputs. This model has been accessible on all paid plans since September 2024.
Industry Collaborations: In September 2024, Runway partnered with Lionsgate and launched The Hundred Film Fund, committing at least $5 million to support AI-augmented film projects.
Funding and Valuation: Runway has attracted significant investment, raising $141 million in June 2023 from prominent backers such as Google, Nvidia, and Salesforce Ventures, leading to a valuation of $1.5 billion.
Google VS OpenAI VS Runway.💪💪💪💪
— X Dog (@AIDog1984)
5:42 PM • Dec 25, 2024
🎥 While OpenAI and Google are making big strides in AI video generation, Runway's emphasis on professional-grade tools and industry partnerships highlights its strategy to stand out by tailoring it’s offerings to active film industry creatives.
Meta Platforms is set to launch a standalone AI application, Meta AI, in the second quarter of 2025. This move aims to position Meta competitively alongside AI offerings from companies like OpenAI and Google.

Integration and Expansion: Currently, Meta AI is embedded within platforms such as Facebook, Instagram, WhatsApp, and Messenger. The forthcoming standalone app is designed to extend its reach beyond these platforms, providing users with a dedicated AI experience.
Enhanced Capabilities: Meta AI offers functionalities including answering questions, generating images, and editing photos. Recent updates have introduced memory features to deliver more personalized recommendations.
Monetization Strategy: Meta is exploring a subscription-based model for the app, aligning with industry trends where premium AI services offer advanced features for a fee.
Investment in AI: CEO Mark Zuckerberg has announced plans to invest up to $65 billion to advance Meta's AI initiatives, underscoring the company's commitment to leading in the AI sector.
JUST IN: Meta, $META, to release standalone Meta AI app in effort to compete with OpenAI's ChatGPT, per CNBC
— unusual_whales (@unusual_whales)
9:19 PM • Feb 27, 2025
🤖 Is anyone going to care if Meta’s AI has it’s own app? Late comers at this stage, Meta will probably rely on the prevalence of it’s other apps (Facebook, Whatsapp) to drive adoption — similar to Elon Musk’s xAI using X as one big advertising platform for Grok.
🆕 Updates
Pika 2.2 is HERE! 🎥✨
What’s new?
✅ 10s video generations for longer, more dynamic clips
✅ 1080p resolution for sharper, high-quality visuals
✅ Pikaframes – Key frame transitions anywhere from 1-10s for more creative control— Wes Roth (@WesRothMoney)
7:00 PM • Feb 27, 2025
HOLY SHITT, Microsoft dropped an open-source Multimodal (supports Audio, Vision and Text) Phi 4 - MIT licensed! 🔥
> Beats Gemini 2.0 Flash, GPT4o, Whisper, SeamlessM4T v2
> Models on Hugging Face hub, integrated with/ Transformers!Phi-4-Multimodal:
> Modalities: Integrates… x.com/i/web/status/1…
— Vaibhav (VB) Srivastav (@reach_vb)
5:53 AM • Feb 27, 2025
Exa Websets is now live! 🚀
Supercharged search awaits. 👇
— Exa (@ExaAILabs)
5:58 PM • Feb 25, 2025
Say hello to Ideogram 2a, our fastest and most affordable text-to-image model to date -- optimized for graphic design and photography.
Now live on the Ideogram website, API, and partner platforms for all users.
— Ideogram (@ideogram_ai)
5:02 PM • Feb 27, 2025
📽️ Daily Demo
🚀 Rendernet just launched Clothes Swap!
Now you can change your AI characters' outfits in seconds👗
Perfect for AI influencers, eCommerce, and more to showcase endless fashion possibilities
🧵Here's how it works
🔖Bookmark for later
— MayorkingAI (@MayorKingAI)
6:52 PM • Feb 26, 2025
🗣️ Discourse
𝗢𝗽𝗲𝗻𝗔𝗜 𝗶𝘀 𝗶𝗻 𝘀𝗲𝗿𝗶𝗼𝘂𝘀 𝘁𝗿𝗼𝘂𝗯𝗹𝗲
They still have the brand name, a lot of data and tons of (mostly unpaid) users, but:
• GPT 4.5 is hugely expensive
• Even so it offers no decisive advantage over competitors; zero moat.
• Scaling hasn’t gotten them to AGI… x.com/i/web/status/1…— Gary Marcus (@GaryMarcus)
2:28 AM • Feb 28, 2025
i've been testing gpt 4.5 for the past few weeks.
it's the first model that can actually write.
this is literally the midjourney-moment for writing.
(comparison to gpt 4o below) x.com/i/web/status/1…
— ben (@benhylak)
8:39 PM • Feb 27, 2025
deepseek r1 release: open source o1
grok 3 release: beats every benchmark
gpt 4.5 release: Can hold my hand when I am scared— samsja (@samsja19)
7:25 PM • Feb 27, 2025
Important update: Figure is launching robots into the home
Our AI, Helix, is advancing faster than any of us anticipated, accelerating our timeline into the home
Therefore, we've moved-up our home timeline by 2 years; starting Alpha testing this year
— Brett Adcock (@adcock_brett)
6:13 PM • Feb 27, 2025