🤖 OpenAI Releases “GPT-4.5”

Subscribe | AI Toolkit | Meet The Team

Welcome back AI enthusiasts!

In today’s Daily Report:

• ⚙️OpenAI Releases “GPT-4.5”

• 🧠LLMs “Think” Like Developers When Coding

• 🤯New dLLMs Generate Over a 1,000 Tokens per Second

• 🛠Trending Tools

• 🥪Brief Bites

• 💰Funding Frontlines

• 💼Who’s Hiring?

Read Time: 3 minutes

🗞RECENT NEWS

OPENAI

⚙️OpenAI Releases “GPT-4.5”

OpenAI released “GPT-4.5,” the company’s largest and most knowledgeable AI model yet.

Key Details:

• “GPT-4.5” showcases better writing capabilities, improved world knowledge, and what OpenAI calls a “refined personality.”

• OpenAI CEO Sam Altman explained that it’s the first AI model that “feels like talking to a thoughtful person to me.”

• OpenAI warned that “GPT-4.5” isn’t a frontier AI model, but it’s “OpenAI’s largest LLM.”

• It was fine-tuned using Reinforcement Learning From Human Feedback (RLHF), which uses human feedback to teach LLMs to self-learn more efficiently and align with human preferences.

• It’s currently available to ChatGPT Pro Plans and developers across all Paid API Tiers, with ChatGPT Plus, Team, and Enterprise Plans getting access next week.

Why It’s Important:

• “GPT-4.5” is super expensive to run, with developers across all Paid API Tiers paying 30x the input cost and 15x the output cost to use it.

• “We’re out of GPUs,” said Altman. “We’ll add tens of thousands of GPUs next week…This isn’t how we want to operate, but it’s hard to perfectly predict growth surges that lead to GPU shortages.”

🩺 PULSE CHECK

AI RESEARCH

🧠LLMs “Think” Like Developers When Coding

Meta’s FAIR Team developed “SWE-RL,” which enhances the reasoning abilities of LLMs to help them tackle real-world coding tasks.

Key Details:

• “SWE-RL” relies on Reinforcement Learning (RL), which teaches LLMs to learn the optimal behavior in an environment to obtain the maximum reward.

• RL is comprised of four components:

  1. Learner: The LLMs

  2. Environment: The real-world coding tasks the LLMs interact with.

  3. Policy: The instructions the LLMs follow to take action.

  4. Feedback: The Positive Rewards or Negative Penalties the LLMs observe after taking action.

• Positive Rewards are given when the LLMs successfully write, debug, and test code.

• Negative Penalties are given when the LLMs generate inefficient, error-prone code.

Why It’s Important:

• When developers tackle real-world coding tasks, it’s not just about writing code; it’s about testing, debugging, and refactoring that code over and over.

• “SWE-RL” helps LLMs not only generate code but also “think” like developers by constantly taking in feedback to adjust code.

INCEPTION LABS

🤯New dLLMs Generate Over a 1,000 Tokens per Second

Inception Labs developed “Mercury,” a family of diffusion Large Language Models (dLLMs) that generate text faster than ever.

Traditional LLMs generate text from left to right, one Token at a time. In other words, a Token can’t be generated until all the text that comes before it has been generated.

Tokens are the smallest units of data used by LLMs to process and generate text. Similarly, we break down sentences into words or characters. You can think of Tokens as syllables. Simply put, Tokens represent bits of raw data; a million tokens equals roughly 750,000 words.

Instead of generating one Token at a time, dLLMs generate entire blocks of Tokens in parallel for increased speed, efficiency, and control. It’s 10x faster than traditional LLMs, 10x cheaper than traditional LLMs, and 2x the size of traditional LLMs with the same latency and cost.

🛠TRENDING TOOLS

🎓DeepTutor is your personalized AI tutor.

📬Forage Mail declutters your email inbox.

📽️Topaz Labs brings old videos back to life.

👷Basalt integrates AI into your product in seconds.

🍬getcaramel.ai turns your ideas into revenue-generating ads.

🔮Browse our always Up-To-Date AI Tools Database.

🥪BRIEF BITES

Hugging Face launched “FastRTC,” an open-source Python library that helps developers build real-time audio and video AI apps.

Vevo Therapeutics created “Tahoe-100M,” the world’s largest single-cell dataset that maps out 60,000 drug-cell interactions.

NVIDIA CEO Jensen Huang said that nearly everyone would benefit from having a personalized AI-powered tutor with them at all times.

IBM unveiled “Granite 3.2,” a family of small AI models that deploy Conditional Reasoning, Time Series Forecasting, and Document Vision to tackle Enterprise workloads.

💰FUNDING FRONTLINES

• Hyperlume closes a $12.5M Seed Round to transform AI Data Center Connectivity.

• Bridgetown Research secures a $19M Series A to build AI Agents for Enterprise Research.

• Variational AI raises a $5.5M Seed Extension for AI-driven Small Molecule Drug Discovery.

💼WHO’S HIRING?

• Trepp (New York, NY): Data Science Intern, Summer 2025

• CACI (Sterling, VA): Software Engineering Intern, Summer 2025

• K2 Space (Los Angeles, CA): Loads and Dynamics Engineer, Entry-Level

• Advarra (Remote): Data Scientist, Mid-Level

• ThoughtSpot (Mountain View, CA): Senior Staff AI Architect, Senior-Level

📒FINAL NOTE

FEEDBACK

❤️TAIP Review of The Day

REFER & EARN

🎉Your Friends Learn, You Earn!

You currently have 0 referrals, only 1 away from receiving ⚙️Ultimate Prompt Engineering Guide.

Copy and paste this link to friends: https://theaipulse.beehiiv.com/subscribe?ref=PLACEHOLDER