Alejandro Peire

Business e Industrial Controller - Economista - Máster Project Management.
Mi perfil de Twitter se centra en mi trabajo y actividad en mercados financieros.

[Alejandro Peire @AlexPeire]

@FindleysFinance @VJNCapital
What avg. growth would you assign to each of these companies for 10/15 years of P/E or P/OCF?
If the growth you consider is higher than the number it says in the picture in the time horizon selected, it is undervalued.
I would appreciate your opinion

[Alejandro Peire @AlexPeire]

@ASM65617010 @pepmartorell Hace 100 años ese estudiante para hacer hallazgos iba a usa papel y lápiz. Hace 5 años iba a usar un ordenador muy avanzado, se iba a apoyar en él. Mañana ese estudiante se apoyará en su IA. La IA es un compañero de trabajo brutal, un socio. El estudiante tendrá que buscar socio

[Alejandro Peire @AlexPeire]

@VJNCapital @FindleysFinance Hmmm, I would prefer $GOOG over $AMZN, but $GOOG & $META are too correlated, almost same business.
I would choose $MSFT or $UBER over $GOOG and $AMZN.

[Alejandro Peire @AlexPeire]

@GarethRansome @EMostaque Another advice I could give you is that you connect both codex and claude to GitHub (free) and make them chat and share there. Make them do collaboration there. Create a "Jira" kind of PM there for them to work together.
That's another thing that has been working very well to me.

[Alejandro Peire @AlexPeire]

@GarethRansome @EMostaque I get good results (not always...) by working always with codex+Claude code. They work together, plan together. They are much better than me planning the other one's job and correct it. With this strategy is working quite well to me.
But I still am in the learning process

[Alejandro Peire @AlexPeire]

@GarethRansome @EMostaque Our brain has the same problem and bigger. So we learn to methods cut problems into pieces and organize the solving. We didn't learned enough of how to do it with AI brains. AI brains are not magical. We have to cut problems too in AI and eat them bite to bite, with strategy.

[Alejandro Peire @AlexPeire]

@oguzerkan That is because you may not have tested Fable 5.1. it is absolutely incredible.

[Alejandro Peire @AlexPeire]

@richardhutton @witheleanorai Careful, supposedly cache is much cheaper in fable 5.1. maybe the long context doesn't count.
Go see how many output tokens has eaten. In my case 5h window was 4.6M tokens. Not minutes but tokens eaten like crazy

[Alejandro Peire @AlexPeire]

@ApexSMK @AnthropicAI Max 5x se ha fundido limite de 5h en 1 disparo, 1:30h. Ahora bien... En un solo disparo ha hecho algo que Opus o Sol habrían necesitado 20 iteraciones. En 1:30h, 1 shot, ha gastado 4,6M de tokens de salida para hacer algo inimaginable. En 1 shot he acabado el trabajo de 2 días...

[Alejandro Peire @AlexPeire]

@bridgemindai Same here but with a gigantic difference. In one shot has made flawlessly what 10 or 20 shots in Sol or Opus wouldn't. The output quality is simply ludicrous, crazy, and those word don't cover it.
Our reasoning maybe should be work once and go to sleep for 4h with job done.

[Alejandro Peire @AlexPeire]

@sonypictures_es OK. I have tried Fable 5.1 and it is fu**ing ludicrous. It made the impossible in 1 shot. Never seen something like it, not even close... But this MONSTER ate the 5h usage in 1 prompt and it didn't even finished the job with Max 5x.
This eats a lot!!!!

[Alejandro Peire @AlexPeire]

@AnatoliKopadze I could agree that Opus 5 was smart, but for the last 2 weeks is really dumb, really really dumb... It was my favorite model and I don't know WTF have they done to it. Simple boring tasks screwed without reason. Today every task, every single one, was completelly flawed, useless

[Alejandro Peire @AlexPeire]

@thsottiaux OK! I have to admit it. @AnthropicAI has nerfed Opus 5 till make it stupid. Today it is screwing up absolutelly everything, even simple tasks. And @OpenAI is reducing credit allowence, so basically they are screwing with our workload. I will have to try chinese apparently

[Alejandro Peire @AlexPeire]

@chetaslua Well, today Sol High eats all my credit allowance in 30 minutes. So with this fantastic Astra I suppose I will have 1/2 of a prompt.

[Alejandro Peire @AlexPeire]

@kimmonismus @0xdominus If half the US starts losing weight rapidly with this drug the global warming will go through the roof with all that CO2 liberated

[Alejandro Peire @AlexPeire]

@buildwithrajath Yes, but probably the stupidity in the models lately comes from Anthropic training those new models. They must have nerfed them all to get enough compute to train.

[Alejandro Peire @AlexPeire]

@JonhernandezIA They are thinking in terms of LLM not in a mechanism where the LLM is part of it. That's probably where Gemini is losing the AI race and the reason it feels so stupid using it.

[Alejandro Peire @AlexPeire]

@jp54362 If you surpass 400k context window you are screw now a days in Claude, since 1 month aprox. They must be training something new. But don't let Codex trick you, it is not better. It could be in certain situations. I am going back to Claude...

[Alejandro Peire @AlexPeire]

@buildwithrajath I am really close to work between Claude that gives me much more and Chinese models for workhorse job lead and guided by Claude.
GPT in codex gives me only headaches

[Alejandro Peire @AlexPeire]

@jpschroeder Another thing is that Anthropic models get really stupid once they reach 400k context window. I start to sweat when closing to it. They are better and tend to do thing really great untill 400k but then they start to f*ck everything up. GPT are 10x better at context compaction