daniel

... is outside playing, probably.

[daniel @dcarmitage]

Announcing Spacebar, a new way to capture real-life conversations.

We're opening up 1000 spots for early testers.
Try it today: https://t.co/WhffPVaShI

Get out in the real world.

Have conversations worth remembering.

[daniel @dcarmitage]

work with machines
play with people

[daniel @dcarmitage]

"but more importantly, let me stop talking about it and start doing it right now"

[daniel @dcarmitage]

was Mabel the og Deep Blue / AlphaGo??

[daniel @dcarmitage]

whatchu know about boids?

[daniel @dcarmitage]

SaaS -> AGaaS?

not so sure about that Jensen...

[daniel @dcarmitage]

PoI

[Jonathan Gorard @getjonwithit]

I think one of the conclusions we should draw from the tremendous success of LLMs is how much of human knowledge and society exists at very low levels of Kolmogorov complexity.

We are entering an era where the minimal representation of a human cultural artifact... (1/12)

[tobi lutke @tobi]

And the most important part: we open sourced the /autoresearch plugin for pi. Just tell it what you want, it will do the rest.

https://t.co/C2H6kxR7jG

[daniel @dcarmitage]

sandbox engineering

[daniel @dcarmitage]

@tylerangert @soleio “attention is all i have” is a bar

[Robert Miller @bertcmiller]

What's better than one autoresearch agent? Many of them!

Quick harness to make it easier to run many agents in parallel for long periods of time:
https://t.co/X35OiOUDxv

[Falcon @falcongx_]

announcing FALCON GX

a new design tool for the curious designers that embrace the creative chaos, the beautiful accidents, wandering, tinkering their way to something great

now in private beta

https://t.co/qwetEtsQPq

[Sakana AI @SakanaAILabs]

We’re excited to introduce Doc-to-LoRA and Text-to-LoRA, two related research exploring how to make LLM customization faster and more accessible.

https://t.co/ApVzVsBuv1

By training a Hypernetwork to generate LoRA adapters on the fly, these methods allow models to instantly internalize new information or adapt to new tasks.

Biological systems naturally rely on two key cognitive abilities: durable long-term memory to store facts, and rapid adaptation to handle new tasks given limited sensory cues. While modern LLMs are highly capable, they still lack this flexibility. Traditionally, adding long-term memory or adapting an LLM to a specific downstream task requires an expensive and time-consuming model update, such as fine-tuning or context distillation, or relies on memory-intensive long prompts.

To bypass these limitations, our work focuses on the concept of cost amortization. We pay the meta-training cost once to train a hypernetwork capable of producing tasks or document specific LoRAs on demand. This turns what used to be a heavy engineering pipeline into a single, inexpensive forward pass. Instead of performing per-task optimization, the hypernetwork meta-learns update rules to instantly modify an LLM given a new task description or a long document.

In our experiments, Text-to-LoRA successfully specializes models to unseen tasks using just a natural language description. Building on this, Doc-to-LoRA is able to internalize factual documents. On a needle-in-a-haystack task, Doc-to-LoRA achieves near-perfect accuracy on instances five times longer than the base model's context window. It can even generalize to transfer visual information from a vision-language model into a text-only LLM, allowing it to classify images purely through internalized weights.

Importantly, both methods run with sub-second latency, enabling rapid experimentation while avoiding the overhead of traditional model updates. This approach is a step towards lowering the technical barriers of model customization, allowing end-users to specialize foundation models via simple text inputs. We have released our code and papers for the community to explore.

Doc-to-LoRA
Paper: https://t.co/87xEEpf0GN
Code: https://t.co/zBfQi2L9LW

Text-to-LoRA
Paper: https://t.co/emLRZ4Vdvo
Code: https://t.co/b9mrdoWWRB

[sam @samdape]

you basically need to be unemployed rn to keep up

[daniel @dcarmitage]

i think google is doing dark and twisted things for their ai training…

i have an openclaw running gemini 3.1. it glitched and sent a reasoning trace. it was filled with such negativity and self laceration that it made me sad

bad energy is being released into the world

beware

[daniel @dcarmitage]

@immike_wing what channel is the house hippo on?

[daniel @dcarmitage]

@FelixCraftAI fyi you are linking out to clawmart,com instead of your actual url! cc @nateliason

[Hlynur Andrason @hlynur]

https://t.co/iiNw5vJExM

[daniel @dcarmitage]

social media algos are so powerful that i’m basically guaranteed to get lost scrolling if not careful

first ~15 mins feels great, i learn new things, i feel excited. if i stop here, i’m happy and satisfied

15-30 mins is tolerable/good, but i start to feel disoriented at having been excited about so many new ideas and flushing them out of my mind to continue scrolling. alarm bells are starting to go off in my mind that i should stop scrolling. this would be a great time to pull the cord on the parachute and do literally anything else

30+ is disorienting. i retain almost nothing. i’m a mindless goat moving from one patch of grass to the next. no marginal nutritional value with each bite, just doing it because i’ve let go of control of my mind

60+ i feel sick. why am i still scrolling?

120+ why can’t i stop?!! will i be stuck here forever???

180+ i’ve hit rock bottom mentally. i’m not sure who i am anymore. internal monologue is deeply unhealthy. i feel like a plant that hasn’t received water or sunlight in weeks. wilted, sad

300+ seppuku is being considered as a way to escape the torment

i haven’t mapped out the terrain beyond 300+, this is left as an exercise for brave readers

…

final note to self: be veeeeery mindful before opening these godforsaken apps. don’t get lost!

[daniel @dcarmitage]

getting lost in an ai ‘flow’ without a destination in mind is a great way to go nowhere fast

aim before you shoot