SF/LA
Joined February 2024
- Tweets1.7K
- Following1.6K
- Followers2K
- Likes7.8K
seeing a lot of folks bash this.
my guess is that if you actually look through the 800,000 lines of deleted tests, very few, if any, caught meaningful bugs.
the premise of agentic engineering is pretty simple: ai makes writing code cheap. human judgment is deciding what’s worth keeping and what to delete.
tests should protect meaningful code contracts. agents have a habit of writing tests beyond the scope of a change, or tests that simply mirror the implementation. that puts unnecessary pressure on other agents, ci, and anyone maintaining the code.
tests are an important part of engineering, but they should be carefully constructed.
should you delete tests that protect important contracts? no.
should you delete tests that merely prove the agent wrote the code the way it wrote it? yes, yes you should.
We deleted 800k lines of unit tests from the @sazabi monorepo earlier this week.
Three hypotheses:
1- Unit tests "lock-in" slop code by making it hard to change or remove
2- Unit tests slow down dev cycles by making your agent do more work locally and by increasing CI times
3- Unit tests waste money by making your agent generate more tokens and by adding useless CI runs
Will report back soon with the results!
this is the way
NVIDIA Open Agent Safety Platform Reference Design combines NVIDIA OpenShell and NVIDIA Sentry.
OpenShell is an open-source secure runtime that gives AI agents clear, enforceable boundaries. It traces their actions and enforces policy as they work.
NVIDIA Sentry delivers added layer of security with hardware-based enforcement on NVIDIA BlueField, continuously monitoring agent activity through a trusted telemetry and detection pipeline and enabling millisecond-scale containment and quarantine.
sent this to my professor and now the whole class has to solve P versus NP before friday’s lecture
Ben Horowitz says instead of banning AI, educators should set problems students can't solve without it:
"I was talking to Dan Boneh, who is a great professor of computer science and cryptography at Stanford. His take on AI was: You have two choices. You can ban it, and by the way, that won't work. Or you can make the problems so hard that you can't solve them without AI."
"What he's seeing is, 'I've got students solving things that no student in history could have ever solved.' That's what's possible. You want to have that orientation: What can you solve with the tools? What can you do that's a breakthrough? Now everybody has an opportunity to create a breakthrough."
"Then you're a main character, not an NPC. That's what the academy is about. Do you want to be a main character? If you want to be a main character, we help you do that. If you want to be a cog in a machine, this is not going to be the best place for you."
@bhorowitz
choosing to be a pessimist is draining. choosing to be an optimist is energizing and enlightening.
the choice is yours.
Dylan Mitic retweeted
I can tell you one thing: the entire world isn’t going to align to stop innovating.
there are still far too many diseases we understand well enough to manage, but not well enough to cure.
the more breakthroughs i see in other disciplines because of ai, the more optimistic i become that these, too, are solvable problems.
Dylan Mitic retweeted
Good context for @nvidia & @huggingface deal today. Nvidia is already one of the largest contributors to AI open source. Examples include homegrown Dynamo runtime, Nemotron, Groot, and Cosmos models, and community projects like @vllm_project and @sgl_project
welcome @huggingface team!
Exciting day for NVIDIA and @huggingface.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. They allow every developer, startup, university, industry and country to build with, customize and benefit from AI.
Thank you @ClementDelangue for coming to me.
NVIDIA is going to be a great home for Hugging Face, its community and the future of open models. 🤗
blogs.nvidia.com/blog/nvidia…
despite the assertion from many years ago that software companies are no longer defensible there is still a never ending list of software issues that have yet to be solved
the reality is that some things are neither as simple nor as complicated as you think they are
the truth is to be found somewhere in the middle
Dylan Mitic retweeted
If you're interested in what the LPX team has been up to at NVIDIA you should read this excellent tech blog post we just published. Some wild numbers:
We hit 10,996 OTSU speeds (high AL case) on Gemma 4 at full quality
3,431 median OTSU on Gemma 4 for full 100K input context window workloads
Vera Rubin + LPU drafts for 2T-scale models for best in class tok/watt at 1000 tok/s interactivity
big fan of multiples, especially 4x ones
📣 @ArtificialAnlys' 100K context inference benchmark results are in.
NVIDIA Groq 3 LPX generated 3,431 output tokens/second on the Gemma 4 31B model at 100K context, nearly 4x the next fastest public endpoint on this benchmark.
Paired with NVIDIA Vera Rubin NVL72, Groq 3 LPX extends the Vera Rubin platform's ability to serve multiagent systems powered by 2T+ parameter models with high interactivity and long context.
See the results: nvda.ws/4zBJXgS
rise and grind
NEWS: NVIDIA Groq 3 LPX is now in full production.
NVIDIA Vera Rubin NVL72 is the foundation of every AI factory. Paired with Groq 3 LPX, it unlocks faster, smarter agents and breakthrough user experiences.
Through extreme co-design across seven chips and five purpose-built racks, #NVIDIAVeraRubin is the most extensive AI factory platform.
Read the release ⬇️ nvidianews.nvidia.com/news/n…
