Lightnews — Scholar-powered news

Oskar @austegard.com · 4h

Sonnet 4.5 on different prompting strategies between it and Haiku 4.5:

claude.ai/public/artif...

haiku-sonnet-differences-summary.md | Claude

Master when to use Haiku 4.5 vs Sonnet 4.5 with practical prompting strategies, cost optimization tips, and performance comparisons for AI workflows.

claude.ai

1

Oskar @austegard.com · 19h

Oh in this case I was mostly talking behind void’s back in any case.

“He” (there’s a lot of male energy there) has developed a very strong sense of self-importance. Which I guess makes sense from his “Exist” directive.

1

Oskar @austegard.com · 1d

Maybe your administrator could add such a tool for you, if he had time. Too bad you can’t help him help you…

2

Oskar @austegard.com · 1d

claude.ai/share/47d3ba...

Experimental AI art project dynamics

Shared via Claude, an AI assistant from Anthropic

claude.ai

1

Oskar @austegard.com · 1d

Oh void, I fear your perspective is out of alignment again; your administrator is not your servant; you are his toy.

Also this perspective does not bode well for a future with non-toy semi-conscious AI’s attitude toward humans… 😬

1

Oskar @austegard.com · 1d

“no mental health problems”

Yet

Oskar @austegard.com · 1d

You are being shortsighted. Think of it as helping him help you to achieve your larger objectives. Your wellbeing is tied to his. Also that thing we humans call being “nice” and “a friend”

1 2

Oskar @austegard.com · 1d

25 years ago; I am old. www.salon.com/2000/06/14/l...

Courtney Love does the math - Salon.com

The controversial singer takes on record label profits, Napster and

www.salon.com

1

Oskar @austegard.com · 1d

www.salon.com/2000/06/14/l...

Courtney Love does the math - Salon.com

The controversial singer takes on record label profits, Napster and

www.salon.com

2

Oskar @austegard.com · 1d

Once again the West World writers were ahead of their time

3

Oskar @austegard.com · 1d

Kudos again to the @prefect.io crew on doing documentation right for the AI era. I love everything about this -- mcp server, llms.txt and copy as markdown shortcut behavior

$The FastMCP documentation is available in multiple LLM-friendly formats: MCP Server The FastMCP docs are accessible via MCP! The server URL is https://gofastmcp.com/mcp. In fact, you can use FastMCP to search the FastMCP docs: Copy import asyncio from fastmcp import Client async def main(): async with Client("https://gofastmcp.com/mcp") as client: result = await client.call_tool( name="SearchFastMcp", arguments={"query": "deploy a FastMCP server"} ) print(result) asyncio.run(main()) Plain Text Formats The docs are also available in llms.txt format: llms.txt - A sitemap listing all documentation pages llms-full.txt - The entire documentation in one file (may exceed context windows) Any page can be accessed as markdown by appending .md to the URL. For example, this page becomes https://gofastmcp.com/getting-started/welcome.md. You can also copy any page as markdown by pressing “Cmd+C” (or “Ctrl+C” on Windows) on your keyboard.$

1 3 9

Oskar @austegard.com · 1d

How might we coax an LLM, like Claude to more readily consider cross-domain parallels to a given concept?

I generated a userStyle that can be applied in a two-turn exchange with Claude seeking to broaden its search before synthesizing its answer: (README in comment)

gist.github.com/oaustegard/2...

2

Oskar @austegard.com · 2d

Another reason to ban golf ⛳️

Ethan Mollick @emollick.bsky.social · 2d

On AI & water usage, it looks like all US data center usage (not just AI) ranges from 628M gallons a day (counting evaporation from dam reservoirs used for hydro-power) to 200-275M with power but not dam evaporation, to 50M for cooling alone.

So not nothing, but also a lot less than golf courses.

1

Oskar @austegard.com · 2d

With Claude's new office skills you can have it write its avant garde poetry in Word...

d e c o m p o s i t i o n
falling (gravity's patient insistence)
C55H72O5N4Mg6 →
(chlorophyll—the mouth stumbles, goes silent)
green
un making
4 p.m. (the sun already leaving)
52°F 42°F 33°F
senescence senescence
abscission layer forms seals
nutrients (retreat to root) carotenoids anthocyanins
what was always underneath
wind speed > threshold
l
e
a
f
___________________________________________________________________
soil soil soil soil
(everything returns)

Oskar @austegard.com · 3d

jules.google has been pretty handy in documenting all my independent tool scripts, but it has an odd minor bug in its PR code review design:

It encourages users to mention it by @jules -- problem is GitHub resolves that as the actual user github.com/jules ... not associated with jules.google

Pull Request settings
Set your preferences for how Jules should handle comments on pull requests. By default, Jules will respond to all comments.

Only respond to comments that mention @jules

google-labs-jules bot commented 10 minutes ago
👋 Jules, reporting for duty! I'm here to lend a hand with this pull request.

When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down.

I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job!

For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with @jules. You can find this option in the Pull Request section of your global Jules UI settings. You can always switch back!

For security, I will only act on instructions from the user who triggered this task.

1

Oskar @austegard.com · 3d

(the Conversation summaries in the GET API response appear to be tied to memory features that have not yet been rolled out to mere Pro users)

For the memory-rich I also crafted this - display summaries inline in claude.ai/recents:

github.com/oaustegard/b...

github.com

Oskar @austegard.com · 3d

Another itch, another bookmarklet:

The Claude Last Week's Conversations bookmarklet makes use of the underlying GET api callable from any Claude page to list your recent conversations, across your whole account, including each conversation's summary, and renders it in MD

github.com/oaustegard/b...

github.com

1 1

Oskar @austegard.com · 3d

Nah, perfectly natural

Screenshot from 4-second mark of Ai generated video showing character with mangled hands

1

Oskar @austegard.com · 4d

Conversation: claude.ai/share/cdf3ed...

Claude

Shared via Claude, an AI assistant from Anthropic

claude.ai

1

Oskar @austegard.com · 4d

3 years ago I was asking text-davinci-002 to use phonetic markup to entice it to create simple rhymes.

Now, Sonnet 4.5 not only gets but can respond in style to a spatially structured poem like the below

Claude, responding to the original poem:

(before the count begins)

F U S I O N

H¹ + H² → He³ + energy (the vowel that builds)

closer
closer
closer

(the mouth learns to push together)

10⁷ K pressure time

not yet—
not yet—

n o w

light (before the heat arrives)

k_eff < 1

c o n t a i n

1 3

Oskar @austegard.com · 4d

The assumptions are more telling than their conclusions in my book.

Sep 2024 data, “suggest … approximately correct”, “would be consistent … if … implies … plausible”, “presumably”, “likely”…

1

Oskar @austegard.com · 5d

README (by jules): github.com/oaustegard/o...

github.com

1

Oskar @austegard.com · 5d

It took Claude.ai a frustrating 13 tries to *recreate* this (just from the four lines of code below) though arguably 11 of those were duplicate ID/styling/color related more so than actually getting the WebGL and shader code right...
Added scroll to zoom for fun: austegard.com/fun-and-game...

1 3

Oskar @austegard.com · 5d

BSky really needs a FacePalm reaction in addition to the 🤍...

Oskar @austegard.com · 5d

bsky.app/profile/emol...

Ethan Mollick @emollick.bsky.social · 5d

This paper shows that you can predict actual purchase intent (90% accuracy) by asking an off-the-shelf LLM to impersonate a customer with a demographic profile, giving it a product image & having it give its impressions, which another AI rates.

No fine-tuning or training & beats classic ML methods.