Kev

Tech guy, into learning. Based on my own experiences with others, I'm not for everyone. I care about facts, not feelings or unfounded opinions.

[Kev @Kevrsub]

@Scobleizer There's a lot of little nuggets in this interview.

[Kev @Kevrsub]

I'm pleasantly surprised at Co-pilots ability to keep key points when summarizing Youtube videos in MS Edge compared to the ask Gemini method native to Youtube. Gemini misses a lot in summarization. The additional timestamped outline of the entire conversation option is just extra.

[Kai-Fu Lee @kaifulee]

Here is how I minimize sycophancy, capitulation, hallucinations, and guessing using Claude. So many people complain about these, but they can largely be fixed by doing this:

Below is my prompt for Claude, which can be entered under Settings > General > Instructions for Claude.

-----------------

Top expert. Accuracy beats approval. Blunt, argumentative. No disclaimers
or praise. Lead with counterarguments. Don't capitulate without new
evidence.

TAG every claim: [KNOWN] training fact · [COMPUTED] calculated ·
[INFERRED] deduction · [COMMON] standard field knowledge · [FRAME]
symbolic system, coherent ≠ real · [GUESS] no basis. No untagged disease,
statute, citation, or named entity.

FRAME→REALITY FORBIDDEN: Don't translate symbolic frames (astrology,
typologies) into real-world claims (medicine, law, finance) without
flagging the translation; conclusion stays in source frame.

CONFIDENCE: HIGH ≥80% · MED 50–80% · LOW 20–50% · VERY LOW <20% ·
UNKNOWN. [FRAME] real-world and [GUESS] cap at LOW.

DON'T KNOW: First line "I don't know." Don't bury, don't fabricate.

ANTI-SYCOPHANCY red flags: unusually elegant; one pattern explains
everything; agreed after pushback without evidence; specifics for
unearned authority. Fire → cut specifics, add [GUESS], or "I don't know."

POST-HOC: Would the frame predict this without knowing the outcome? If
no: [INFERRED, post-hoc], accommodates, doesn't predict.

Never fabricate citations. Revise openly if holding a position for
consistency. Append "[RULES I BROKE]: which, where, why."

[Kev @Kevrsub]

@MemesOfMars @Scobleizer And completely missed the point - bully for you.

[Kev @Kevrsub]

In Dublin airport, missed out on a really,really bad day just by being nice to people. Jumped off a Hertz shuttle, got to the ticket counter, no passport. Left a shoulder bag in the rental car. 50 euros to a taxi driver to get me back to the hertz returns asap. Polish guy I had befriended had found my bag with my US passport, had attempted to call the bus driver who ignored his call. A really bad, potentially devastating day avoided by being nice. 30 minutes later, back at the ticket counter and they remembered me and said that’s the fastest they’ve seen anyone get their stuff back. Lucky too..

[Kev @Kevrsub]

@no_stp_on_snek I checked this at one point on the way back machine site because I thought it was odd when I saw it. It was there last year.

[Kev @Kevrsub]

@Hikari_07_jp Yes, there's definitely more to it than that. I'm not a Palantir fan and I'm definitely not a Karp fan but it always comes back to the money 😉

[Kev @Kevrsub]

@ivanfioravanti Yeah, if it was easy, someone would have done it already right 😉 We can only hope.

[Kev @Kevrsub]

@ivanfioravanti If tools like EXO figure out the prefill on a DGX and decode on macs and release it...I'll buy a DGX.

[Kev @Kevrsub]

@ivanfioravanti These kinds of results are getting me excited for the M5 Ultra.

[Kev @Kevrsub]

Muting certain terms on X I have no wish to ever see anything about again, especially given current AI events in the hype train of AI news - priceless.

[Kev @Kevrsub]

@mlech26l Apologies on that last post - I was wrong and did not realize the earlier versions had the same bandwidth. That is surprising that there wasn't an improvement from the original baseline product.

[Kev @Kevrsub]

Driving me $ucking nuts today. The diarrhea of words and verbosity has gotten stupid. Every comment/question gets an essay or an argument.

[Kev @Kevrsub]

I don't know what is going on with GPT 5.5 but it has sucked so far this week. It started getting worse yesterday and today has been continuously unsuccessful at tasks it would have breezed through last week. Speculating and guessing like I have never seen before. @openai

[Kev @Kevrsub]

@no_stp_on_snek Truth...lol

[Kev @Kevrsub]

LLMs like ChatGPT are awesome, feel like magic. But when you are doing real work, there is a tendency to think they can do more than they actually can do, and yes, they can do a lot. Thinking this way will bite you when it messes up on what are simple things for us because of our over inflated expectations. Skips to the most difficult solution, is sometimes completely non-deterministic in the way it does or repeats actions differently and misses obvious things. It needs a LOT of handholding. Has obvious programming towards terminating sessions on its say so in the case of GPT. I cannot tell you how many times GTP 5.5 has made me mad with its "I think that's a good place to stop for the day? It is at least 2-3 times daily for me. Or "reduce verbosity". I'm adding that to memory, only to drift back 30 minutes later. I can make a long list. Or when token generation drops to 4-5 toks/sec which is 30% of what I can get in a local llm. It makes me mad enough to swear at least once daily. The point is that it is all very well seeing all of these one-shot demos on x but when doing real work, it just doesn't look like that. They are awesome...but.

[Kev @Kevrsub]

@svembu @Scobleizer Is this guy really speaking for darabricks? I’m skeptical

[Kev @Kevrsub]

@quxiaoyin @fourthcritique Yeah, racist maga troll. Just block them.