dhruv

Art x Code x Weirdness(?!) #creativeAI #generative #AIart Thread (1/n)

@mhdempsey Building something similar based on all the photos in my phone from art galleries. Would love to try this out and give feedback!

DM me if you want to test and give feedback on a new thing i built around finding art that evokes the feelings of other art you love (think blended goodreads/letterboxd/spotify)

@willccbb miss python tho

like if you’re even kinda good at kubernetes or C++ you’re unbelievably employable rn

fake software engineering jobs are dying rapidly. real software engineering jobs, however,

@bcherny @karpathy how do you balance learning something vs executing in this mode? in the case of memory leak, do you then ask it to summarise + list down tools to help debug it easier for future? (e.g. if someone doesn’t know what profiler options to use, asking Claude Code to write it up)

I feel this way most weeks tbh. Sometimes I start approaching a problem manually, and have to remind myself “claude can probably do this”. Recently we were debugging a memory leak in Claude Code, and I started approaching it the old fashioned way: connecting a profiler, using the app, pausing the profiler, manually looking through heap allocations. My coworker was looking at the same issue, and just asked Claude to make a heap dump, then read the dump to look for retained objects that probably shouldn’t be there; Claude 1-shotted it and put up a PR. The same thing happens most weeks. In a way, newer coworkers and even new grads that don’t make all sorts of assumptions about what the model can and can’t do — legacy memories formed when using old models — are able to use the model most effectively. It takes significant mental work to re-adjust to what the model can do every month or two, as models continue to become better and better at coding and engineering. The last month was my first month as an engineer that I didn’t open an IDE at all. Opus 4.5 wrote around 200 PRs, every single line. Software engineering is radically changing, and the hardest part even for early adopters and practitioners like us is to continue to re-adjust our expectations. And this is *still* just the beginning.

I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.

@andrey_kurenkov Jeff Dean talks about it a bit here as well youtu.be/AnTw_t21ayE?t=…

wow never knew this "The Google Brain project began in 2011 as a part-time research collaboration between Google fellow Jeff Dean and Google Researcher Greg Corrado. Google Brain started as a Google X project and became so successful that it was graduated back to Google." Google getting into AI research in 2011 (before AlexNet / deep learning!) with Andrew Ng was really ahead of its time in hindsight! en.wikipedia.org/wiki/Google_Br…

@soorajchandran_ its also useful to go for photo walks without a camera - just notice and take mental pictures of what you would have clicked. mull over it. then go back again with a camera

If you want to get better at photography, just slow down a bit. Most people point their phone and click without thinking. Instead, take 5 seconds to actually look at the frame. Do you even like what you’re seeing? What’s the subject? Where’s the light coming from? Is the lighting bad? And after you take the shot, spend 10 seconds looking at it and critiquing it. You can learn rules like the leading lines, rule of thirds etc, later. First, just learn to pause and care.

@StasBekman @huggingface Ran into this a few times, but I think setting HF_DATASETS_CACHE / HF_HOME var also solves this?

Just discovered python's `tempdir` silently ignores `TMPDIR` env var if the overridden path doesn't exist. So you're likely to run into `No space left on device` with @huggingface `datasets` if your `/tmp` is smallish as it uses `tempfile` for almost everything. I'm suggesting `datasets` work around this: github.com/huggingface/da…

@chrislakin funnily enough, it was the opposite for me. I could feel (pleasure, arm) but could never equate it with happiness imo, both are important

can't believe it took me decades to realize feelings are better described as tuples `(sensation, location)` rather than emotion words

@andimarafioti Is it global RNG for numpy?

I heard X likes ML interview questions. So here's one: what's wrong with this PyTorch DataLoader worker seed?

@gabriberton people need to know about cohen’s kappa

Want to create a difficult test set? Just use wrong labels and no model will even get close to 100% 😞 People who don't understand the importance of inter-annotator agreement should not be allowed to create benchmarks

@cantrell @reve would love it if you could drag the object outside the image bounds and it expands the image

Shipping continues. You can now mirror/flip regions in the @Reve editor. Image arenas are going to need a new category for direct manipulation.

@madiator an ideal solution would be the way we hotswap lora for image models - training a lora per document/set of docs and swapping it during inference as per need

I love how he has articulated what he wants clearly. But people still don't seem to understand what this is. It's not RAG, it's not notebooklm, it's not lmstudio. It's not a solved problem at all, contrary to what anyone claims. I have also always wanted to something like this, but never got to it due to technical hurdles, but also possibly the business angle. But let's aside the business angle for a minute. Please note that I led a Neurips paper called "Generative Retrieval" (arxiv.org/abs/2305.05065) where we transfer the "recommendation knowledge" into a transformer. So that's where my interest in this field comes from. A reasonable start is to do CPT (continual pretraining) of the unstructured text (but needs some data curation) followed by some sort of SFT (and maybe even RL), all of which can be curated. The paper to read here, for the former, is "Synthetic Continual pre-training" from Stanford: arxiv.org/abs/2409.07431 (besides the above one). And it's worth pointing out that ultimately this is first and foremost a data curation problem, and then a modeling problem. The simplest form of what he is saying is to take a pdf (something that's reasonably long) and then train a llm out of it such that it has internalized the knowledge in the pdf and no RAG is needed. Again, this seems easy but it's not. And then of course there is a lot of fascinating technical problems here, besides the challenges around cost (CPT itself is expensive, but the data gen process -- if you read the above paper -- is also expensive). For example, CPT can make the model a bit dumber in other aspects. So how do you methodically add existing pretraining data (which people don't have access to) during the CPT process.

moody portraits for the moody weather outside

@ku1deep People need to know about perceptually uniform colour palettes

Another terrible map. That gradient is not data it is an opinion.

@kalomaze i feel if you collect your own data, look at the data, and train models, you intuitively guess that hey maybe batch shouldn’t look the same? though maybe easier to grasp when dealing with e.g. computer vision rather than LLM RL

the most painfully neglected "machine learning theory" stuff worth grasping - that actually matters - which most uni courses teaching you ML theory will gloss over - are batch construction rules, the dangers of not shuffling samples/class imbalance, mean reduction biases, etc.

@TheAnnaGat ive one of two girls holding lamps though

Oh gosh there was this beautiful tweet yesterday of a painting of two girls holding candles (Netherlands? Late Renaissance?) Anyone has it? From one of the big art history accounts

@PhadkeTai thanks for the tweet though, went and saw it!

@c_valenzuelab would love a research meetup!

Working from the Runway London offices this week 🇬🇧

postcards from 🇳🇱

not a bad view to wake up to