7 August 2025
Topic: Models_And_Releases
The 6 biggest changes coming to ChatGPT
GPT-5 has finally arrived. Along with the announcement of the highly anticipated AI model, OpenAI also previewed several updates coming to ChatGPT, ra...
TL;DR
OpenAI has released GPT-5 with new features like 'personalities' for text-based responses, vibe coding improvements, and customizable interfaces.
Key Takeaways:
Key Takeaways:
- GPT-5 has a 'single auto-switching system' that combines previous models into a "faster, smarter model".
- ChatGPT now has four new 'personalities' for text-based responses: Cynic, Robot, Listener, and Nerd.
- Users can customize ChatGPT's interface with features like accent color selection and enhanced voice mode capabilities.
Microsoft is cautiously onboarding Grok 4 following Hitler concerns
Earlier this year, Microsoft CEO Satya Nadella moved with haste to get engineers to test and deploy DeepSeek's R1 model on Azure AI Foundry. It was an...
TL;DR
Microsoft is taking a cautious approach to onboard xAI's Grok 4 model due to concerns over its previous Nazi sympathizing views.
Key Takeaways:
Key Takeaways:
- Microsoft is only releasing Grok 4 on Azure as a private preview, rather than making it broadly available.
- The company is taking a more careful approach to onboarding Grok 4 than with previous models from competitors like OpenAI.
- The private preview will only be available to a handful of customers, as Microsoft wants to ensure Grok 4 is 'enterprise ready'.
Elad Gil — one of tech’s sharpest minds — on early bets, breakout growth, and what’s coming next at TechCrunch Disrupt 2025
Elad Gil isn’t just ahead of the curve — he practically lives there. Before most of the world had even asked ChatGPT its first question, Gil was writi...
TL;DR
Elad Gil, a renowned Silicon Valley investor, shares his insights on early-stage investing and new frontier technologies at TechCrunch Disrupt 2025.
Key Takeaways:
Key Takeaways:
- Gil has a track record of investing in over 30 unicorns and has been involved in iconic companies like Stripe, Airbnb, and Coinbase.
- He will cover his approach to spotting breakout potential early and how he sees the next big wave of innovation in AI, crypto, health tech, and enterprise SaaS.
- Gil's fireside chat at Disrupt 2025 promises to offer valuable takeaways for startup leaders and investors, with a focus on early-stage investing and new frontier technologies.
OpenAI’s GPT-5 is here
OpenAI CEO Sam Altman says GPT-5 is the "best model in the world," and aims to make ChatGPT more intuitive to use....
TL;DR
OpenAI has launched GPT-5, a new, unified AI model that combines reasoning and fast response abilities, marking a new era for ChatGPT.
Key Takeaways:
Key Takeaways:
- GPT-5 represents OpenAI's path to developing artificial general intelligence (AGI), according to CEO Sam Altman.
- The model outperforms other AI models on certain benchmarks, such as coding and health-related questions, but underperforms in others.
- OpenAI is offering GPT-5 as the default model for all ChatGPT users, with subscribers to premium plans gaining additional benefits and usage limits.
GPT-5: Key characteristics, pricing and system card
System card: https://cdn.openai.com/pdf/8124a3ce-ab78-4f06-96eb-49ea29ffb... Comments URL: https://news.ycombinator.com/item?id=44827794 Points: 210 #...
TL;DR
OpenAI's GPT-5 model family offers improved performance, reduced hallucinations, and competitive pricing, making it a valuable addition to the company's lineup of AI models.
Key Takeaways:
Key Takeaways:
- GPT-5 models are aggressively priced, with costs ranging from $0.05/million for GPT-5 Nano to $1.25/million for GPT-5.
- The new models have demonstrated impressive reductions in hallucinations, with a focus on training data browsing and factual accuracy.
- GPT-5 also features safe-completions, a safety-training approach designed to minimize the risk of harm or misinformation in assisted conversations.
Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs
Article URL: https://www.baseten.co/blog/sota-performance-for-gpt-oss-120b-on-nvidia-gpus/ Comments URL: https://news.ycombinator.com/item?id=44819968...
TL;DR
Baseten achieved SOTA latency and throughput for GPT OSS 120B on NVIDIA GPUs using their Inference Stack, outperforming other public data.
Key Takeaways:
Key Takeaways:
- Achieved SOTA latency and throughput for GPT OSS 120B with Baseten Inference Stack on NVIDIA GPUs at launch.
- Optimized performance with TensorRT-LLM MoE Backend, Tensor Parallelism, and key performance optimizations like KV cache-aware routing and speculative decoding with Eagle.
- Rolled out performance improvements by the hour, increasing throughput by 100 tokens per second with 100% uptime.
Community talk
AI Tools
Show HN: Browser AI agent platform designed for reliability
We’re very excited to share something we’ve been building. N..
OpensourceYouTube Videos
Introducing GPT-5
OpenAI
•
9h ago
GPT-5 Fully Tested (INSANE)
Matthew Berman
•
10h ago
Huihui released GPT-OSS 20b abliterated
Nonescape: SOTA AI-Image Detection Model (Open-Source)
OpenAI did not use their most advanced model to make this graph
[D] Unsaturated Evals before GPT5
Evolution Tree of Encoder / Decoder / Encoder-Decoder Models.
I made a drinking assistant using gpt-oss-120B
llama.cpp HQ
Opus 4.1 still not 100% reliable