Puja Abbassi (@puja108) header
Puja Abbassi (@puja108) profile picture

VP Product @GiantSwarm; CNCF Ambassador; CKA; recovering Data Science Academic; into Chinese Martial Arts; not updating his website; @puja@hachyderm.io

Cologne, Germany
Joined April 2009
Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)Photo by Puja Abbassi (@puja108)
This looks promising
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor. Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon. thinkingmachines.ai/news/ink…
2
105
That is crazy fast.
رسميا BYD U9x اسرع سيارة على وجهة الارض بسرعة وصلت ٥٠٢ كيلومتر في الساعة
1
1
162
Bonkers!
ok this is just ridiculous
79
This is pretty awesome, fine tuning using a frontier model seems to be super efficient.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2
94
Puja Abbassi retweeted
Leider geil 😂😂😂😂
72
477
84
2,239
85,899
Puja Abbassi retweeted
TIL that global spending on alcohol is still larger than AI infrastructure spending this year ($1.7T vs $1.5T)
70
123
91
2,687
779,373
Puja Abbassi retweeted
Today we're standing up XBOX LABS (labs.xbox.com). Here is the note that I sent to the team: Team, The first video games emerged some 75 years ago. In the time since, there have been booms and busts, but every year there are more great games to play, more devices to play them, and best of all, more people than ever playing. That growth helped make XBOX what it is today, and we've helped shape the industry in return, from bringing hard drives to consoles and online gaming through Xbox Live to Game Pass and cloud gaming, alongside franchises that have brought millions of people together. In recent weeks, we've seen games rapidly decompiled and rebuilt to run on hardware they were never originally designed to support. Windows PC builds have been translated to run on Apple Silicon or ported to run inside a browser through WebAssembly and WebGPU, the latter now approaching 90% global browser support. Perhaps more surprisingly, we've seen many of the biggest games in the world have been remixed into the source code of another. These experiments challenge long-held assumptions about our industry: the tether between software and hardware, the boundary between creation and play, and how people discover and experience games. While it is hard to predict the future of gaming, it is easy to see the possibilities in making great games easier to access, giving old games new life and allowing more players to become creators and co-creators. These are unmistakably good outcomes for players, even when (and sometimes because) they challenge our existing business models. To explore these possibilities, I’ve asked Scott to stand-up XBOX LABS, a small group of game makers, engineers and scientists whose job is to turn emerging ideas into things our players and developers can experience. We'll build in the open, learn from what people do, and measure success by whether more people are playing and creating. I think XBOX has an important role to play here. Coding became the first major workload for agents because they can write, run, and test software. Games share many of those same properties, and people spend more than 72 billion hours a year playing XBOX games, roughly five times the estimated time spent writing code worldwide. We've spent 25 years learning how to make great games, and I want our game makers to help shape what these new technologies make possible. As we learn, we’ll share the tools we build with developers across the industry, just as we always have. Our priorities aren't changing. We remain committed to the 4Cs, to making great games, to our hardware and to being responsible stewards of IP and the communities built around it. LABS gives us a way to explore new technology that is game design led without abandoning the work already underway. Asha
1,730
3,585
1,343
40,764
5,167,040
Puja Abbassi retweeted
Some mistakes you might be making: Your effort level is too high Your skills are too long You're not aligning before you start work You're aligning too often, and too hard Your automated checks are too lax Your custom tooling is too token-inefficient You aren't doing enough automated review You aren't reviewing previous coding sessions (with an agent) You're doing too much human review You aren't vibe coding enough prototypes Your agent can't verify its own work You're babysitting the agent too much You've overloaded your global steering files Your codebase is hard to navigate You aren't building a shared language Your codebase architecture allows bad tests to be written You aren't thinking strategically enough You aren't thinking enough
273
247
60
4,022
211,850
Puja Abbassi retweeted
I’ve had a few people write to ask me to follow up on a (reply!) tweet I wrote that’s gotten a lot of retweets. I considered deleting it just to avoid the headache, but I think that would make people more paranoid. So let me share my thoughts.
Replying to @kmad @matthew_pines
I think we might lose public key cryptography.
48
298
36
1,906
269,416
This is very close to what I've built for us internally, will need to try it out, but it still feels focused on personal vs organizational memory, which is rather what I'm focusing on in a memory consolidation phase
Introducing Dreaming: Across sessions, Devin builds a memory graph of how you like to work At night Devin self-improves its memory to remove stale records and discover latent information We are creating an OSS standard called Agent Memory Repo: cognition.com/agent-memory-r…
104
Very much this! There's also good mcp and skill support by now for home assistant. And it's very hackable, I had a built in integration break fot my old AC so I had Claude patch an out of tree one and now run my own fork of that until upstream agrees on a solution
I’d suggest using home assistant since it has support for 99% of everything and if you don’t like the support just patch it versus hand rolling your own interface Personally I use home assistant and then the home assistant APIs to pipe into custom interfaces
2
130
Been thinking about exactly this recently. We could not establish good PaaS-like abstractions in platform engineering as even golden and paved path models were too restrictive for human devs but agents don't complain and even strive under such limitations
My hot take from chatting to @poteto is that we should use MORE abstractions in the AI age You can use them (combined with harsh lint rules) to reduce the design space available to the agent and constrain them only to good decisions. Combined with the fact that high-leverage abstractions let you do more with less code - so, more token efficient. Plus, unwinding the damage from a bad abstraction is much cheaper with agents. This runs counter to a lot of folks thinking that agents just want to read the raw code. They can, but they're not maximally efficient that way. Be braver! Design abstractions.
2
113
Most people’s use cases are not asking for Tibet or Tiananmen. Theres tons of professional use cases including. coding, operations, data science, and more that these models save money and foster sovereignty. And if you worry, tune them to be uncensored or get the pre-built ones
113
Puja Abbassi retweeted
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
1,705
6,740
1,566
56,610
8,173,517
Finally
Claude Code will now try to find a graceful stopping point when you hit your 5-hour limit mid-task, instead of cutting off mid-edit. It gets a small, fixed allowance pulled from your weekly limit to wrap up what it can.
3
234
Puja Abbassi retweeted
Look people We don't have enough cloud sandboxes. It's that simple. > One per developer! < That's the target. Each dev must correspond to one sandbox product.
The sandbox announcements with effectively identical and wildly undifferentiated technology will continue until morale improves.
8
3
2
8
4,260
This looks super interesting and geniuinely different
This quoted post is unavailable.
2
159
Nice one!
When several people talk at once, a transcript can get messy fast. Our new Nemotron 3 Diarization model tracks who spoke when, even when voices overlap. It handles up to eight speakers, has 100M parameters, and is now available on @huggingface 🤗
103
This is cool
🥳 Excited to start revealing what we've been working on in the last few months. First, we decided to reinvent Kubernetes for agentic workloads with statefulness and fast resumption. Secondly, we are building an agentic orchestrator that will be Google's open agentic orchestrator and runtime. github.com/google/ax
143
Free live stream FROM AGNTCON + MCPCon EU now for the next 36 hours 🙌
1
115