“When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker.“
This came from a piece by Margaret-Anne Storey on "cognitive debt", picked up by @simonw, and which I personally also encountered. While building a synth engine from scratch, I kinda lost the plot. Then I let Claude document each "phase", and from that I generated Remotion videos. It worked quite well, and I learned a lot, much more than I would have from just reading the source code!
It's on Lovable so you can just fork it and take a look (the prompt is in supabase/functions/transform-render/index.ts):
Transform this idealized architectural rendering into the most brutally realistic, depressing photograph possible. This is the WORST CASE scenario - what the building will actually look like in reality:
- Set on a dreary, grey, overcast late November day with flat, lifeless lighting
- The sky is a uniform dirty grey, threatening rain
- All trees are completely bare - just skeletal branches against the grey sky
- The landscaping is dead, muddy, or non-existent. No lush gardens, just patchy brown grass and bare dirt
- Remove ALL people, the scene should feel empty and abandoned
- Any water features should look stagnant and grey
- Add realistic weathering, dirt streaks, and construction residue on the building
- The building materials should look how they actually appear, not the idealized clean version
- Include visible utility boxes, drainage grates, and other mundane infrastructure usually hidden in renders
- The overall mood should be bleak but realistic - this is what buyers will actually see on a random Tuesday in late autumn
- Maintain the exact building, angle, and composition, just strip away all the marketing polish
The goal is honest truth, not beauty. Show what the architect's client will actually see when they visit the site.
Hmm, it seems Claude Research is only available for the Max plan:
> Research is now available in early beta for Max, Team, and Enterprise plans in the United States, Japan, and Brazil. Simply toggle on the Research setting in chat.
Janelle Shane’s work has been really inspiring. Her talk at Strange Loop from a couple of years ago — right before DALL-E, before ChatGPT — was a turning point, and very funny:
As a HyperCard fan I was struck by this quote from Bill Atkinson:
“I have realized over time that I missed the mark with HyperCard,” he said from his studio in Menlo Park, California. “I grew up in a box-centric culture at Apple. If I’d grown up in a network-centric culture, like Sun, HyperCard might have been the first Web browser. My blind spot at Apple prevented me from making HyperCard the first Web browser.”
It failed my simple "fox, goose, corn" benchmark. Here's the prompt:
"Farmer to cross a river with fox, goose and corn in a boat that takes at most two. Fox eats goose and goose eats corn if left alone.
Farmer can row the boat himself and doesn't need a boatman to ferry them across.
How can he cross the river with all the three unharmed in least number of trips?"
It starts promising but fails in step 5:
Let's think through the problem step by step:
1. The farmer needs to transport three items across the river: a fox, a goose, and some corn.
2. He can only take one item with him on each trip because if he leaves the fox and the goose together, the fox will eat the goose, and if he leaves the goose and the corn together, the goose will eat the corn.
3. Therefore, the first trip should be with the goose since it is harmless and won't cause any damage to the other items.
4. After dropping off the goose at the other side of the river, the farmer returns to pick up the fox for the second trip.
5. Once the fox is dropped off at the other side, the farmer returns again to pick up the corn for the third and final trip.
6. Upon reaching the other side, the farmer reunites with his unharmed fox, goose, and corn.
Final answer: It takes 3 trips to cross the river with all three unharmed.
I tried it out myself (git pull && make) and the difference in results are day and night! It's amazing to play with, although you should prompt it differently than ChatGPT (more like the GPT-3 API).
SDL is not a GUI library: it doesn't have any buttons, sliders or text controls. That's what this library provides.
To get on something on screen, it provides a number of backends, including SDL. However there is also one that talks to X11 directly, providing the minimum dependencies needed on Linux (as far as I'm aware).
Basically the library generates a list of drawing commands. It is up to the underlying backend (SDL, GLFW, X11, your own embedded one,....) to draw these on-screen.
“When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker.“