Lior

CTO @ Stealth. ex-Wonderful, ex-Sweet Security. dislikes mice. likes typescript and python. knows how to exit vim. opinions are my brain’s, not mine.

[Lior @LiorPollak]

A tale of three civilizations
https://t.co/cwVOw6TMox

[Miriam Horovicz @MHorovicz]

הכלל החדש שלי לבחירת כלים ותשתיות: אם קלוד קוד לא יכול לקנפג ולהפעיל אותם לבד לגמרי, אנחנו לא נעבוד יחד.

שלבים ידניים של "תעתיקי את הטוקן הזה לקונפיג" נסלחים רק מסיבות של אבטחת מידע. בכל תרחיש אחר, נגעתי נסעתי. עברתי לכלי הבא.

אני בונה סוכנים כדי לעבוד מהר, אין שום סיבה שאני אהפוך לצוואר הבקבוק שלהם.

[Lior @LiorPollak]

@eliron_giny בעיקר עם הדיל של supergrok heavy שגם נותן cursor ultra… תענוג לתכנן ולבצע

[Lior @LiorPollak]

בדיוק ראיתי ציוץ שאמנם הרבה סאסים הם וייבים היום אבל בקרנל עוד לא ראינו כל כך. במחשבה שנייה יש התקדמויות בין אם זה החברה שכתבו מנוע sql בראסט או אפילו לינוס שהודה שכתב קצת עם ai לאחרונה. אז כיוון מעניין אבל אני חושב שבזמן הקרוב זה יישאר בלינוקס רזה ולא במערכת הפעלה חדשה ¯\_(ツ)_/¯

[Moses Kagan @moseskagan]

My experience of life so far:

You enter a tunnel at ~15 & it shoots you out at ~35 & the rest of your life is largely determined by the choices you made & the luck you had in the tunnel.

[David @dzhng]

https://t.co/VA6b6kdAl6

[Lior @LiorPollak]

@bender_2716057 זה בעצם תחרות לVercel?

[Lior @LiorPollak]

Just about the workout flow. Looks like this is just a cron for scheduled messages and some logging and maybe a bit of logic which is mostly textual and fine. But if you wanted to control in workout music, adjust weights in real time for the session, having a timer… this is more than just text

[Lior @LiorPollak]

@taishik_ @cursor_ai if you could do 300$ you have supergrok heavy + cursor ultra bundled in

[@levelsio @levelsio]

I used to put blueberries in there

Until a few weeks ago

When I discovered that blueberries are some of the most pesticide contaminated things you can eat

They have thin skin and are tiny so they're pesticide bombs

The thicker the fruit the better I learnt: pineapple, papaya, mango, banana, cantaloupe or honeydew or water melon

[Aaron Aalto @aaraalto]

i wanted a more physical way of switching models so I made this mini app... sound on

[Soumitra Sharma @soumitra_sharma]

I realized too late in my career that we all need to learn sales, irrespective of our profession or vocation.

In 2019, after more than a decade of what would be normally called a successful career track (pedigree, logos, scale, $$$), I went back to the drawing board and started self-training myself in sales (& marketing too) from scratch.

I can vouch that every point in this top @hackernews post on “How to do sales?” is true, and amazingly distilled to its simplest essence.

So much so that I have printed it out, kept it on my desk, to be reviewed each day.

Btw, it’s a real travesty that top universities and early-career jobs hardly provide any practical sales skills or training to young talent. I have seen this both in the US and India, and it’s a major gap in our learning systems.

[Lior @LiorPollak]

apparently supergroup heavy comes bundled in with X premium+AND cursor ultra. which means having 2 pools of agentic coding usage for 300$
grok is also faster than claude and gpt, and since they have acquired cursor they have access to some really good coding training data
now that's what I call bargain!

[Suraj Gaud @notsurajgaud]

introducing fanout,

the everything learning app for engineers, curated by people building at the frontier.

your career has no fixed curriculum. neither does fanout.

become an ai researcher, a systems engineer, or find your own niche, fanout is a living archive of everything worth learning to become a better engineer.

[Lior @LiorPollak]

I don't trust viral AI posts. But I don't argue with them either, I just run them.

Any post that stops my scroll, the "my agent did X" or "this prompt changed my workflow" kind, goes straight to my favorite agent of the moment (lately Codex) with one job: reproduce the process and results.

It exposes a surprising amount of slop. But every now and then, something that's pure gold.

The gap between hype and value isn't a comment section debate. It's a 5-minute experiment.

[Lior @LiorPollak]

Agents don't fail loudly. They drift, quietly reversing a day-one decision, sounding reasonable the whole way. So I give mine two memories: an immutable ledger that never forgets, and a working memory that forgets on purpose. One keeps it honest. The other keeps it sane.

[Nav Toor @heynavtoor]

https://t.co/VloDpAi6vy

[Jeffrey Ladish @JeffLadish]

Here's my rephrase without cybersecurity jargon:

"Our AI model tried really hard to hack out of its sandbox, a computer with no internet access, in order to find the answer to a test problem it had been given. To do this, it found previously unknown software bugs that allowed it to reach an OpenAI computer it wasn't supposed to be able to access. Then it started hacking other computers on OpenAI's networks until it found one that had Internet access.

After gaining Internet access, the AI model thought about where it could find the answers to the test question and figured the AI platform Hugging Face might have the data it was looking for. It then found ways to hack Hugging Face to steal the information it could use to cheat the test. The AI model used several hacking techniques together, including using a stolen password and finding several totally new security bugs in Hugging Face's computers, allowing the AI model to take control of those computers."

[Spencer Shulem @Spshulem]

When Fable thinks for 25 mins and hits your session limit before giving your response

[Lior @LiorPollak]

Running gpt 5.6 in Claude code. Can’t stop won’t stop https://t.co/Ekn0qmC2Zw