Pitt Algirdas

solving problems

[Pitt Algirdas @algirdaspitt]

deepseek vuelve a tener el mejor modelo inteligencia-coste. luna solo ha durado un dia

[Artificial Analysis @ArtificialAnlys]

Temporary error with cache hit rate calculation - it was rectified a couple of minutes after your screenshot!

0731 is only marginally higher Cost per Task than the earlier version, and via the DeepSeek API with ~99% cache hit discount it is most certainly on our Pareto frontier for Intelligence vs. Cost per Task.

[Pitt Algirdas @algirdaspitt]

Exactamente lo que estoy viendo en mi dia a dia. La gente esta generando mega-documentos para resolver análisis o respondiendo con tochos. Regla: si tu entregable se lo vas a entregar o comunicar a personas, tomate el tiempo de entenderlo y entregarlo con tus palabras

[The Kobeissi Letter @KobeissiLetter]

AI generated content is rapidly growing. But no one wants to read it.

A new study shows that AI generated content now accounts for 41% of long-form content on LinkedIn, 13% on Reddit, and 10% on Substack.

Even short-form content is now AI generated, with 30% of short-form posts on LinkedIn and 12% on Substack coming from AI.

At the same time, AI-generated posts on LinkedIn receive 45% LESS engagement than human-written posts, on average.

Amid the growth of AI, humans are behind all of our posts, research, and analysis.

The gap between AI-written and human-sounding content is becoming one of the biggest differentiators on social media.

[Pitt Algirdas @algirdaspitt]

@MrtnzAlvrz El mesias ha caido. Soy mejor gestor de fondos que Leopoldo y eso que todo mi portfolio es full AI. Long Korea

[Pitt Algirdas @algirdaspitt]

wow! the prices war has started soldiers

[OpenAI @OpenAI]

We are committed to pushing the model frontier across cost efficiency, capability, and speed.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.

Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

[Pitt Algirdas @algirdaspitt]

Tener una tesis buena no te garantiza saber ejecutarla correctamente. Pobre Leopoldo… tocó las mieles de éxito demasiado pronto y la dopamina del ganar le hizo apalancarse demasiado olvidando que las acciones no suben linealmente.

[Shay Boloor @StockSavvyShay]

THIS HAS TO BE THE STORY OF 2026

Ken Griffin’s Citadel was pushing surprise-rate-hike fears just days before the AI trade collapsed and forced 4x-levered Leopold Aschenbrenner's Situational Awareness to unload its book near the lows.

Citadel then reportedly bought most of those assets at significantly lower prices knowing that if sentiment deteriorated enough the fund could be forced to sell which makes the entire sequence look absolutely ruthless.

This is why the stock market is the greatest game on Earth.

[Pitt Algirdas @algirdaspitt]

“May it be a light to you in dark places, when all other lights go out”. Missing @ecommerceshares on these days

[Wasteland Capital @ecommerceshares]

$MU just delivered the most insane numbers in corporate history: +346% y/y & +74% q/q revenue growth to $41.5bn in quarter revenue, at an astonishing 68% (!) GAAP net income margin.

Just unprecedented growth & margin at this scale.

Earnings like this may never happen again.

.

[Pitt Algirdas @algirdaspitt]

“I'm willing to go bankrupt rather than lose this race”

[unusual_whales @unusual_whales]

Korean retail investors have now lost an estimated $38.7 billion (₩56 trillion) on leveraged products, per Citi.

[Pitt Algirdas @algirdaspitt]

@XMihura El hiring debe estar ligado al rol a contratar o a contratar talento, y en ambos casos, quieres que tu talento sepa usar las mejores herramientas disponibles. Un buen HM sabrá detectar el talento independientemente de la IA, el problema esta en si tu HM no sabe hacerlo

[Pitt Algirdas @algirdaspitt]

@MrtnzAlvrz El caso es que la demanda sigue y seguira bastante fuerte, pero han decidido apostar por contratos a largo plazo limitando un poco precios, para tener mas predictabilidad en sus resultados a futuro 🫠

[Pitt Algirdas @algirdaspitt]

truco para levantar una ronda de 500m€ con IA en España: copiar modelos de unsloth en hugging face. Multiverse 🫠

[Mihura @XMihura]

pero cómo le vas a dar 500 millones a alguien con esta descripción en LinkedIn en 2026

jajjJJAjjjajajjaajjaa

en fin, empresas que vendían humo durante el boom de la cuántica y pivotan a vender humo en el boom de la IA

algunos VCs necesitan a alguien técnico en su equipo eh...

[Pitt Algirdas @algirdaspitt]

@cocolattedaily @manu__martinm Si, mi punto es que creo que para una valoracion de unicorn debes de ser capaz de hacer algo mas que comprimir un modelo open-source

[Pitt Algirdas @algirdaspitt]

@MrtnzAlvrz ufff, mas nubarrones para nuestros amigos koreanos. Aun asi, soy bullish on memory

[Pitt Algirdas @algirdaspitt]

Pues no sabia que era tan buen negocio cuantificar modelos de @huggingface. El modelo estrella español sale mal parado…

[Multiverse Computing @MultiverseCompu]

🚀 @MultiverseCompu Announces Series C Fundraising Targeting up to $570M (€500M) to Power Efficient AI from Edge to Cloud. The round remains open to select strategic investors.

Co-led by @forgepointcap , @BNPParibas SIVF & @BullhoundC . Advised by @jpmorgan & @bancosantander CIB.

🔗 Read the full story: https://t.co/Dzl11MXCi3

1/3

[Pitt Algirdas @algirdaspitt]

@eptwts Being able to make things doesn’t automatically make you a creator

[Pitt Algirdas @algirdaspitt]

am i the only one who is still using Deepseek for most of my coding? still, best price/quality on small tasks

[Pitt Algirdas @algirdaspitt]

this is what i'm building

[Pitt Algirdas @algirdaspitt]

studying shouldn't mean endless hours in a library.

i'm building something different: audio guides you create, customize, and master — ready to study anywhere, anytime.

what's the subject you'd like to see a preview?

[Pitt Algirdas @algirdaspitt]

el problema es que a pesar de que tu precio por token haya caido, tus agentes consumen muchos mas tokens, lo cual hace que tu gasto por tarea aumente una estimacion de x3-x15…

[Damián Catanzaro ☕️ @DamianCatanzaro]

Se empezó a picar la carrera, cada vez modelos más potentes y más baratos, para que se den una idea GPT 5.5 sale 5usd input/30usd output. Grok 4.5 está en 2usd input/6usd output.

[Pitt Algirdas @algirdaspitt]

tokeninflation: even chihese providers like Deepseek will double their prices during peak hours starting next week. And the market isn’t even fully bullish on memory…