Shoopy

research over speculation.
AI • Crypto • Automation

[Shoopy @0xShoopy]

Opus 5 just made Opus 4.8 look broken.

Same benchmark, ARC-AGI-3. Opus 4.8 scored 1.5%. Opus 5, out today, scored 30.2%. One generation, a 20x leap.

And it's not a one-off:
→ novel problem-solving (ARC-AGI-3): 1.5% to 30.2%
→ agentic terminal coding (Frontier-Bench): 21.1% to 43.3%, nearly double
→ computer use (OSWorld 2.0): 55.7% to 70.6%
→ business workflows (AutomationBench): 17.0% to 26.0%

Same price as Opus 4.8. Near Fable 5 intelligence at half Fable's price.
Now the part nobody else will tell you: it isn't first at everything. GPT-5.6 still takes agentic coding on DeepSWE (72.7 vs 68.8), and Fable 5 edges it on legal and health.

Read the system card, not the hot takes.

[Shoopy @0xShoopy]

@hitu_monke bookmarked

[Shoopy @0xShoopy]

Opus 5 just made Opus 4.8 look broken.

Same benchmark, ARC-AGI-3. Opus 4.8 scored 1.5%. Opus 5, out today, scored 30.2%. One generation, a 20x leap.

And it's not a one-off:
→ novel problem-solving (ARC-AGI-3): 1.5% to 30.2%
→ agentic terminal coding (Frontier-Bench): 21.1% to 43.3%, nearly double
→ computer use (OSWorld 2.0): 55.7% to 70.6%
→ business workflows (AutomationBench): 17.0% to 26.0%

Same price as Opus 4.8. Near Fable 5 intelligence at half Fable's price.
Now the part nobody else will tell you: it isn't first at everything. GPT-5.6 still takes agentic coding on DeepSWE (72.7 vs 68.8), and Fable 5 edges it on legal and health.

Read the system card, not the hot takes.

[Claude @claudeai]

Introducing Claude Opus 5.

It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price.

[Shoopy @0xShoopy]

Jason Liu just showed how he runs his entire job at OpenAI inside Codex, hands behind his back.

His AIE workshop, in 5 timestamps:

3:37 – compaction works now, so one pinned thread can run for five weeks
6:08 – threads message each other, so you manage a team instead of an IC
9:44 – install plugins for Slack, Gmail, Notion so the agent reads your day
16:06 – appshots hand the agent an app's accessibility tree, not just a screenshot
40:14 – slash-goal loops keep running as long as a verifier says not done

The setup: a personal monorepo of skills and a memory vault, woken up by heartbeats and by other threads.

He dictates with a foot pedal. 75 min, more useful than it sounds.

[Shoopy @0xShoopy]

Jason Liu just showed how he runs his entire job at OpenAI inside Codex, hands behind his back.

His AIE workshop, in 5 timestamps:

3:37 – compaction works now, so one pinned thread can run for five weeks
6:08 – threads message each other, so you manage a team instead of an IC
9:44 – install plugins for Slack, Gmail, Notion so the agent reads your day
16:06 – appshots hand the agent an app's accessibility tree, not just a screenshot
40:14 – slash-goal loops keep running as long as a verifier says not done

The setup: a personal monorepo of skills and a memory vault, woken up by heartbeats and by other threads.

He dictates with a foot pedal. 75 min, more useful than it sounds.

[Shoopy @0xShoopy]

Sydney Runkle just laid out the four loops LangChain wraps around every agent they run internally

Her webinar on loop engineering, in 5 timestamps:

4:28 – loop 1: the model and tool loop that does the work
6:36 – loop 2: a grader scoring the result against a rubric
10:16 – loop 3: triggers from slack, github, your calendar
13:42 – loop 4: an agent reading traces and rewriting the harness
20:01 – where human approval still earns its place

The takeaway: the advantage isn't the agent you build, it's the loops you wrap around it.

45 min, and loop 4 is the part worth staying for.

[Shoopy @0xShoopy]

@0xMovez in the reading pile, thanks man

[Shoopy @0xShoopy]

@choopyplug1 guy spent a decade arguing for scale and now warns about the hype, that's worth listening to bro

[Shoopy @0xShoopy]

Sydney Runkle just laid out the four loops LangChain wraps around every agent they run internally

Her webinar on loop engineering, in 5 timestamps:

4:28 – loop 1: the model and tool loop that does the work
6:36 – loop 2: a grader scoring the result against a rubric
10:16 – loop 3: triggers from slack, github, your calendar
13:42 – loop 4: an agent reading traces and rewriting the harness
20:01 – where human approval still earns its place

The takeaway: the advantage isn't the agent you build, it's the loops you wrap around it.

45 min, and loop 4 is the part worth staying for.

[Shoopy @0xShoopy]

@Yumzlef that's the best outcome honestly, go build it

[Shoopy @0xShoopy]

Sydney Runkle just laid out the four loops LangChain wraps around every agent they run internally

Her webinar on loop engineering, in 5 timestamps:

4:28 – loop 1: the model and tool loop that does the work
6:36 – loop 2: a grader scoring the result against a rubric
10:16 – loop 3: triggers from slack, github, your calendar
13:42 – loop 4: an agent reading traces and rewriting the harness
20:01 – where human approval still earns its place

The takeaway: the advantage isn't the agent you build, it's the loops you wrap around it.

45 min, and loop 4 is the part worth staying for.

[Shoopy @0xShoopy]

Isaac Miller (DSPy) on why a smarter model still won't solve your problem.

in 17 min they lay out how to specify a task so the implementation stops mattering:

→ treat every repeated ai task like a function: fixed inputs, fixed outputs, everything else swappable
→ instructions cover what should happen, in natural language
→ code covers what must happen, because constraints need enforcing, not prompting
→ examples cover what good looks like, for the parts nobody can write down
→ with all three in place the program becomes optimizable, and new techniques drop in without touching your interface

Their closing test for any new technique: hold your prompts, models and code accountable to the problem you actually need solved.

[Shoopy @0xShoopy]

Isaac Miller (DSPy) on why a smarter model still won't solve your problem.

in 17 min they lay out how to specify a task so the implementation stops mattering:

→ treat every repeated ai task like a function: fixed inputs, fixed outputs, everything else swappable
→ instructions cover what should happen, in natural language
→ code covers what must happen, because constraints need enforcing, not prompting
→ examples cover what good looks like, for the parts nobody can write down
→ with all three in place the program becomes optimizable, and new techniques drop in without touching your interface

Their closing test for any new technique: hold your prompts, models and code accountable to the problem you actually need solved.

[Shoopy @0xShoopy]

@LunarResearcher Bookmarked

[Shoopy @0xShoopy]

Isaac Miller (DSPy) on why a smarter model still won't solve your problem.

in 17 min they lay out how to specify a task so the implementation stops mattering:

→ treat every repeated ai task like a function: fixed inputs, fixed outputs, everything else swappable
→ instructions cover what should happen, in natural language
→ code covers what must happen, because constraints need enforcing, not prompting
→ examples cover what good looks like, for the parts nobody can write down
→ with all three in place the program becomes optimizable, and new techniques drop in without touching your interface

Their closing test for any new technique: hold your prompts, models and code accountable to the problem you actually need solved.

[Shoopy @0xShoopy]

Dex Horthy ran the lights-off software factory on his own product, and watched it fall apart.

His AIE talk on why harness engineering isn't enough, in 5 timestamps:

2:20 – this isn't a scale problem, it's a training problem
7:30 – July 2025: nobody read the code, then the site went down
8:56 – models still can't hold codebase quality over time
10:12 – why Claude Code won: the model was trained against its own harness
14:58 – turning the lights back on: product review, architecture, program design

The core of it: RL rewards a passing test, and nothing in that signal punishes bad architecture, because that bill arrives months later.

30 min of planning up front buys back hours of review. 19 min, worth all of them.

[Shoopy @0xShoopy]

@haider1 the shifting deadlines gripe is fair though mate

[Shoopy @0xShoopy]

@VaibhavSisinty k3's genuinely strong, no argument there bro

[Shoopy @0xShoopy]

@intheworldofai everyone's posting a launch today, nobody's posting a link to the announcement, funny that

[Shoopy @0xShoopy]

@birdabo @chetaslua how'd you get access ahead of the announcement chief?

[Shoopy @0xShoopy]

@bridgemindai ping me when the system card is up bro

[Shoopy @0xShoopy]

@alextalksai didn't know this one existed, thx alex!

[Shoopy @0xShoopy]

@0xMovez straight to the watch list man!