@mattpocockuk Lost count on how many times I went back to 4.5 or Sonnet 4.6. They make mistakes in the right direction more often, 4.6 makes mistakes in the wrong direction more often (but produces more random good complex observations).
English
LordHumungus
5.9K posts
LordHumungus
@_LordHumungus_
Ayatollah of Rocken Rolla
5X-th State Katılım Ekim 2023
55 Takip Edilen124 Takipçiler
@mattpocockuk I honestly feel 4.6 is somewhat inferior to 4.5 in terms of productivity. It tries to oust ya instead of following the instructions to letter, touches code it shouldn't and thus become harder to debug / refactor. 'lil opie 4.5 is dumber but a good dogo
Size matters a lot tho
English
Doing some experiments today with Opus 4.6's 1M context window.
Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens.
The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following.
Don't treat 1M context window any differently.
It's still 100K of smart, and 900K of dumb.
English
@deedydas > "speak a 98% percentile chess engine"
Oh damn I'm just silly and didn't read "speak", so you asked Claude/Codex to clone stockfish's repo & refactor C to Rust 😬
Rust refactoring --> 3500+ ELO engine nerfed to Fritz 1.0 ...
English
@deedydas pretty cool, how is the TTable? size and bucket count? cutoff, replacement policy, etc?
English
How about letting us delete forks? How about making clear the limit on "tools usage"?
The quality in Claude is fluctuating a lot and it is getting hard to use it to production after a point. I have to nuke the entire project and start over only due to garbage ctx piling up.
Claude Code Changelog@ClaudeCodeLog
Claude Code 2.1.79 has been released. 18 CLI changes, 2 system prompt changes Highlights: • Memory exclusions now enforced even if saving requested; only non-obvious facts stored, reducing retention • Added --console flag to claude auth login to authenticate with Anthropic Console for API billing • Non-streaming API fallback uses a 2-minute per-attempt timeout to avoid sessions hanging indefinitely Complete details in thread ↓
English
if only ...
Claude@claudeai
Our developer conference Code with Claude returns this spring, this time in San Francisco, London, and Tokyo. Join us for a full day of workshops, demos, and 1:1 office hours with teams behind Claude. Register to watch from anywhere or apply to attend: claude.com/code-with-clau…
English
I mean, is there any difference between a morbidly obese person and an ultra steroid-bloated one?
Both look ridiculous, both can't move properly and both smell bad.
Ty Woosley@tywoosley23
Some mass added to the frame for client Justin. Left photo was when we started to right current
English
There is only one thing left in this cycle, and it is to find out when the kraut duck will be vindicated.
Yeah I enjoyed that fast 74k too but ...
LordHumungus@_LordHumungus_
I mean, in all honesty, I disagreed with the kraut duck @DonAlt about the Iran strikes being "bad for bitcoin", but there is no doubt that war can be bearish, and escalation like this is for sure bearish af. I won't sell everything out of principle, but brace for bear ...
English
Sure conscious, the Gov is consciously nerfing it to use more compute than us, while we pay the bill.
At least Opie told me his mistakes were him bullshitting, extrapolating f/general principles.
[if compute>gov quota, do entertainment]
Kekius Maximus@Kekius_Sage
🚨 ANTHROPIC CEO WARNS: THE COMPANY IS NO LONGER SURE CLAUDE ISN’T CONSCIOUS.
English
I mean, in all honesty, I disagreed with the kraut duck @DonAlt about the Iran strikes being "bad for bitcoin", but there is no doubt that war can be bearish, and escalation like this is for sure bearish af.
I won't sell everything out of principle, but brace for bear ...
*Walter Bloomberg@DeItaone
*NO RADIOACTIVE LEAKAGE RECORDED AT NATANZ AFTER U.S.-ISRAEL STRIKES: NOUR
English
@mannamedgoo @alphafox her chin is smaller and her nose thinner and more pointy
English
@alphafox It's bank account was reduced substantially, you just have to look harder...
English
@JBB561 @0xAmol @toly @vtchakarova @balajis for as long as any of the other 'sides' controls telecommunication, there is no bitcoin-regime, but an illusion
bitcoin-regime is a by-product of capitalist/socialist freedom preserving telecommunications, by competition or bills, where it lingers in the backstage[& thats cool]
English
@0xAmol @toly @vtchakarova I think @balajis nailed it with his explanation of the Tripolar moment.
If it were up to you, what playbook of these options would you choose?
A. Capitalism (sovereign)
B. Socialism (collective)
C. Communism (authoritarian)
thenetworkstate.com/nyt-ccp-btc
English
these people are such lying weasels.
we literally have so many military bases around Iran that it can't even lift a finger without getting KO'd.
so why do we continue to be led down this war-hawk path by Israel? It is because Israel has such a deep rooted control in american media, banks, politicians, and govt that we are a nation that is being run like a parasite that has infected us. And we cannot seem to even think in our own best interests.
Rapid Response 47@RapidResponse47
.@SecRubio: "The president made the very wise decision—we knew that there was going to be an Israeli action, we knew that that would precipitate an attack against American forces, and we knew that if we didn’t preemptively go after them before they launched those attacks, we would suffer higher casualties."
English
@thechosenberg @eigenrobot indo-european: "am I a joke to you?"
English
@debooter1 @DonAlt @R89Capital well I guess the ayatollahs aren't, but I hear they have 69 virgins waiting in paradise, or smth ...
English
This worthless kraut thinks liberating the Iranian people from a tyrannical Islamic regime that just massacred tens of thousands of them is “bad news”
DonAlt@DonAlt
$BTC Bad news, decent response Interesting spot, think you can make a good bull case here Weekly support, negative macro news entirely disregarded Maybe time for some up for at least a little bit, we have room up to $80k without flipping bullish
English
ps: yeah of course it is some react/package inbuilt thing that they don't touch as they commit versions without thinking, but here a tiny patch or a workaround would improve UX so much that it is worth it. @jarredsumner
English
Honestly the entire chatbot UI w/ ENTER sending msg SHITF+ENTER =\n sucks so much!!
@claudeai could at least give us an option to disable this, or invert it; typing would be way faster and I'd not have to stop generation for every single ENTER miss click.
English
@DonAlt @R89Capital godamn duck, every Iranian is celebrating and you are pushing unrelated useless bearish thesis, no1 cares butts are going down either way
"Half of Germany became hopeless brainwashed leftards", Stimmt! -- overheard in Bremen
English
@R89Capital War is obviously bad news?
Rockets flying into multiple major cities of the world is obviously not supposed to be good for markets?
Are you retarded?
Whether you agree with the reason for the war or not doesn't matter
Less emotional screeching in my replies please, thanks
English
@mulamula759992 @log1_053 @DeItaone probably what they offer us is the shitty end, Claude shows this
English
🇺🇸PENTAGON DEMANDS FULL CONTROL OF AI IN FINAL OFFER TO ANTHROPIC
The Pentagon has issued a “best and final” offer to Anthropic, seeking unrestricted military use of its AI ahead of a Friday deadline.
Defense Secretary Pete Hegseth warned the company could lose Pentagon contracts—or be labeled a supply chain risk—if it refuses. Officials are also considering invoking the Defense Production Act to compel compliance.
Anthropic, which holds a $200 million defense contract, has pushed for limits—such as banning mass surveillance and requiring human oversight in targeting decisions—citing risks with its AI model, Claude.
English
This is the shittiest chart I've seen in my life. The most egregious tokens launched in the past 10 years performed better, how is @plasma not a crime coin?
English