Brice
@gdaybrice
Followers
460
Following
130
Media
157
Statuses
2K
Hands-on CTO | Founder | Full-stack AI engineer ⚡️
Sydney
Joined January 2008
Simple visualisation of how gpt-5 routes based on thinking effort compared to previous models.
The level of intelligence for most people in @ChatGPTapp after GPT-5 update will depend highly on the 'router' and how often it would switch into 'thinking' mode. If you were using o3 already, you are unlikely to feel a huge difference. However, vast majority of ChatGPT users
0
0
0
0
0
0
@aisdk Someone asked me which model to choose for tool calling. It's best to evaluate yourself or you can check the Berkeley leaderboard.
0
0
0
Example of agent handoffs (similar to the OpenAI Agents SDK) between a concierge, sales, and refund assistant using a simple transfer-to tool and the @aisdk v5 prepareStep. You can override each step with a different model, system prompt, active tools, and message list.
6
12
137
voyage-context-3 is a new embedding model focusing on chunk-level details with global document context.
blog.voyageai.com
TL;DR – We’re excited to introduce voyage-context-3, a contextualized chunk embedding model that produces vectors for chunks that capture the full document context without any manual metadata…
0
0
1
I like the principles that an agent shouldn't be held accountable (the human in the loop should be) and that it should be clear and transparent about its internal state.
Introducing:.Agent Interaction Guidelines (AIG). Principles and practices for designing agent interactions that integrate more naturally into human workflows. →
0
0
1
Bittersweet memo on the future of design in this AI era. I've always advocated for UX designers to bridge the gap with engineers and get closer to the tools to craft experiences, this is now made easier with AI.
0
0
1
Following @nicoalbanese10 and @__morse suggestions, I adopted this polymorphic schema for saving message parts with the @aisdk
Prefix-based pattern for persisting typed message parts (text and tools) with the @aisdk.
0
0
2
Bet on context engineering over training an end-to-end agentic model, avoid dynamically adding or removing tools mid-iteration, and use the file system as context.
manus.im
This post shares the local optima Manus arrived at through our own "SGD". If you're building your own AI agent, we hope these principles help you converge faster.
0
0
1
Good write-up on how to train an email deep research agent with reinforcement learning to be faster, cheaper, and more accurate than o3.
0
0
0
Looks like @bunjavascript now supports monorepos, no more ghost dependencies 👻.
Bun v1.2.19.- Fixes 163 issues, addressing 1,043 👍.- `bun install --linker=isolated` : pnpm-style isolated symlink installs.- `bun update --interactive`.- `bun why <pkg>`.- VSCode test explorer integration.- Bun.sql gets faster .- `--console-depth`.
1
1
2
I was wondering about the architecture behind the @v0 model, whether it was fine tuned or a hybrid approach. Turns out, it’s both.
vercel.com
Learn how v0's composite AI models combine RAG, frontier LLMs, and AutoFix to build accurate, up-to-date web app code with fewer errors and faster output.
0
0
0
@FanaHOVA Are you using docker only for supabase? Perhaps we can start recommending OrbStack to everyone on macOS. It should only use about 2GB of memory locally.
0
0
1
Voice and browser control are becoming the new normal.
0
0
11