akeynix πŸ”± (@akeynix) header
akeynix πŸ”± (@akeynix) profile picture

ai agents builder, system architecture enjoyor bot dev, data analyst and offchain infra @emperorjournal_ ✨ Ambassador and Atlantean at @AtlantisDEX_xyz πŸ”±

Joined March 2024
Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)Photo by akeynix πŸ”± (@akeynix)
a person with a goal is much more dangerous than one with endless knowledge.
3
105
akeynix πŸ”± retweeted
Your intelligence is capped by your willingness to look stupid while learning.
142
1,473
86
11,828
212,243
its always satisfying when the things you build work as expected. keep buildingggggg
4
165
akeynix πŸ”± retweeted
be so fucking proud of yourself going through it all alone
449
7,581
231
62,996
1,676,115
im grateful to be apart of the tutorial hell, endless stackoverflow solution hunting that fits the exact problem, printing on every line before a statement to see the change of behavior on the script eraaa why? because it shapes me as a technical developer on HOW to properly build things. everyone has ideas, everyone has a vision, a goal. but when you use an agent without an understanding of what you're doing on the low level, that's like building a house without knowing ur bricks are made out of plastics. once you finish and start living in it, a single raindrop causes leakage everywhere, and youre confused as to where tf do i start to fix, only to realize the entire shit is already fucked to ever begin with. so know your shit, dont only build build build. itll bite you hard one day, and youll end up calling for an actual enjinir trying to fix things.
2
2
157
being happy is a true privilege, don't waste it.
2
173
vibe coding really took away the joy of figuring out code logic by head. like there were times where writing a for loop and iterating through data just how we want it to output is so satisfying when it worked. but bcs of it, ideas are so much easier to bring to life now :3
2
3
260
not bad for first time playing MEVs in @RobinhoodCrypto pic only shows wins btw 😹 overall its still -EV bcs its not heavily optimized yet πŸ€“
1
190
opus 5.5 the one shot king is finally back. thanks @claudeai
2
168
is this the "The algorithm is optimizing the AI, and the AI is optimizing the algorithm" moment? πŸ«ͺ
We’re sharing how GLM-5.3 helped build and optimize the inference infrastructure serving GLM-5.3-Flash. The system went from its first successful run to production readiness in less than two weeks, with end-to-end throughput tripling relative to the initial baseline. The key was dense feedback: local correctness tests, execution traces, microbenchmarks, and end-to-end measurements that enabled targeted hypothesis testing rather than reliance on aggregate performance metrics alone. z.ai/blog/glm-built-its-infe…
1
163
akeynix πŸ”± retweeted
how to learn LP? Emperor Journal present: special AMA with @Febzz64 learn how to LP + demo LP with @Invictuslp don't miss it on @emperorjournal_
8
4
33
1,074
okay this shit is terrifying. this model can do RE effortlessly. wtf.
GLM 5.3 FLASH IS THE FKING GOAT. for weeks ive been reading Opus and Fable's lobotomized responses, finally getting out and trying the new GLM 5.3 FLASH brings me back to the times where Opus and Fable were ran at full performance. Thank you @Zai_org ❀️
1
228
GLM 5.3 FLASH IS THE FKING GOAT. for weeks ive been reading Opus and Fable's lobotomized responses, finally getting out and trying the new GLM 5.3 FLASH brings me back to the times where Opus and Fable were ran at full performance. Thank you @Zai_org ❀️
1
1
485
a day in my life using Fable and Opus 5
1
146
akeynix πŸ”± retweeted
Something is seriously wrong with Claude Opus 5 right now. I've been digging through recent user reports, and the same problems keep appearing independently: β†’ bizarre sentence structure and borderline gibberish β†’ unnecessary or invented jargon β†’ hallucinating facts, issues or context that were never there β†’ correcting itself repeatedly inside the same task β†’ ignoring explicit instructions / CLAUDE. md constraints β†’ doing work nobody asked it to do β†’ creating bugs, then burning huge amounts of time fixing its own bugs β†’ declaring tasks finished while leaving major pieces incomplete β†’ turning trivial changes into enormous multi-step projects The communication bug is especially bizarre. People aren't just complaining that it's verbose. They're describing output they genuinely struggle to understand, with mangled phrasing and weird word choices. Basically, one of the smartest coding models on paper occasionally sounds like a drunk senior engineer who swallowed an enterprise software dictionary. Anthropic also recorded degraded-performance incidents affecting Opus 5 on Aug. 17, 18 and 19. Important caveat: that does NOT prove the outages caused these behavioral issues, and I found no evidence that Anthropic intentionally "nerfed" the model. But the complaints are real, widespread, and remarkably consistent. Something is off. @bcherny @AnthropicAI @ClaudeDevs
216
89
36
1,537
77,407
akeynix πŸ”± retweeted
I have cancelled my Anthropic sub until things improve. I do not need my LLM telling me to go to bed, refusing to work because of the remaining context window and other bizarre behavior.
298
270
63
9,024
425,987
did @deepseek_ai launched an update on deepseek v4 flash? seems like this version is much more eager to solve problems than it was yesterday. its less "lazy" with /goal on claude code. rly good shit tho
1
186
is it just me or opus has degraded so bad to a point where it just spams "You're right" almost all the time rn. it used to be able to deduce so many interesting things out of very little info what happened to the one shot king where work is done without much thought @claudeai
3
240
bareng nih, mari pecutkan fable ampe kering
penyakit ketagihan fable 5
4
140