inference engineering // reformed quant
Joined January 2022
- Tweets404
- Following539
- Followers1.4K
- Likes667
flashboy retweeted
Cartesia Managed Agents scores an Elo of 1115 on the Artificial Analysis Speech Agent Arena, 98 points above Cartesia's earlier voice agent product, Cartesia Line, with Task Success Rate up from 77.5% to 91.7%
Managed Agents is @CartesiaAI's fully managed platform for building and deploying voice agents. It is a cascaded system: Cartesia's own Ink Speech to Text and Sonic Text to Speech models run around an LLM of the developer's choice. We evaluated Cartesia Managed Agents with GPT-5.5 as the LLM and the Skylar and Daniel voices.
Key takeaways:
➤ Speech Agent Arena: Cartesia Managed Agents scores an Elo of 1115 (95% CI 1086-1144, 685 samples), 98 points above Cartesia Line at 1017 (95% CI 985-1049)
➤ Agentic tasks: On tasks that need tool calls, Cartesia Managed Agents scores an Elo of 1112
➤ Task Success Rate: 91.7% of eligible tool-calling conversations end with the correct final task-completing tool call
Congrats to the @CartesiaAI team!
See below for more details ⬇️
flashboy retweeted
Ok you’re all going to keep your jobs after all.
flashboy retweeted
i understand why people use partiful but i deeply hate it. only thing it's good at is making people feel like going to your party is going to be a huge pain in the ass and emphasizing how nonessential their attendance is. text your friends and keep an RSVP list in your notes app.
flashboy retweeted
Those in venture, product, media etc. inevitably live in the Nontechnical North
The few engineers here are aesthetes capable of eye contact (spiritually nontechnical)
Tale as old as time: capitalist class in the north, and the oppressed labor class (openai researchers) pushed south
Good software engineers see AI as a tool to build more
Good mathematicians see AI as an extinction event
Tells you a lot about the state of academia
Why is everyone having a meltdown about this? AI didn’t learn some new factoring algo, it ported some code to run on GPUs and ran it on a ridiculous amount of compute
636606729769440499166579950236036751749912014371509557713570027508971809534551913252252094954941974952859310861988904737359709200557919
is a factor of RSA-896
saweis.net/posts/rsa-896.htm…
flashboy retweeted
If all these ex Messari employees had spent as much time building the company as they do now pushing their bags Selkis would have the money to be running Germany
flashboy retweeted
This feels like the day the Bitter Lesson, died. Just scaling compute is no longer the path to easy returns.
The frontier is now algorithmic cleverness, researcher taste and figuring out how to do more with less. No more blindly throwing compute scale at the problem to win
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog: z.ai/blog/glm-5.3-flash
Available now across all official platforms:
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.3…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai
flashboy retweeted
I think we need more clear definitions on AI-assisted coding.
Vibe coding is the “prompt-in product out” form of AI use, where the user just defines an end state (e.g., screenshot or description) and the AI makes all the decisions. Useful and accessible to everyone, but limited in scope of things you can achieve.
Agentic engineering, on the other hand, is when an engineer works step by step to design, implement, refine, and maintain a codebase. More complicated, requires engineering skills, sometimes you have to investigate the code, but can ship scalable software.
