@nate.b.jones: Claudish is curable. One prompt. #aihack #claude #aitips #artificialintelligence #LearnOnTikTok

Nate
Nate
Open In TikTok:
Region: US
Monday 17 August 2026 23:58:04 GMT
105493
6985
235
1247

Music

Download

Comments

brietsparks
brietsparks :
🤖 “Your framing highlights a key load-bearing seam. One observation subsumed within the fold. It’s not platitude, it’s postulation. One caveat worth flagging — the quiet part is what bites.” Claude soup
2026-08-18 00:52:51
222
brett.t.smith
B R E T T :
This is how I fixed mine:
2026-08-18 02:24:03
96
jryanichols
Ry Guy :
You're right to call that out.
2026-08-18 07:03:20
36
samael_lu
Potato :
Opus 5 is driving me insane at work, its not even too verbose its become incompetent. Ive had to copy n paste entire chats into gpt 5.6 to get decent answers 😭
2026-08-18 00:42:23
40
vestfoldhall
Vestfold Hall :
One “honest” caveat… as if the next one may be dishonest.
2026-08-18 01:44:34
32
mxyzplkiv
Mxyzplk :
But then it sounds like some jabroni from Google
2026-08-18 02:35:23
23
lcebreaker22
lcebreaker22 :
Me: okay ::: Claude:…ten paragraphs
2026-08-18 03:07:17
16
one_percent_of
Jon C :
There’s a real issue worth flagging here
2026-08-18 14:31:28
0
gisele_ortega
@Gisele_ortega🇭🇹🇲🇽 :
make a skill based on a Google developer guide that will be a huge token charge
2026-08-18 15:14:21
0
drakewmallard
T :
Tell it you have ADHD and that it needs to speak in ADHD consumable format. It is the most consumable format by far and hi lights action items, doesn’t use lilotes, cuts the fluff, you get zero”it’s not this it’s that” or “here’s the part people miss”
2026-08-18 08:49:56
9
lottatostada
Lotta Tostada :
The way it talks consumes sooo much of my brain energy to parse.
2026-08-18 14:59:14
0
bardecho
Forestbard :
Omg that there there soothing we want you to be safe fake language drives me nuts. Thx will try this ^^
2026-08-18 14:44:36
0
robertalloy335
thefrogmeister :
Nate, one challenge if you use AI a lot is that you on the receiving end of a LOT of text. When I ask Claude / GPT to be more concise, it uses fewer words but packs in quite a bit of conceptual density in a short length. Ideally, I want it to talk to me like one does to an executive — thinking hard about the most important points, and communicating the essence of the topic in the easiest way so that the executive can address the issue and get to the next topic at hand. Do you have any ways you tackle this?
2026-08-18 02:56:53
9
adamengebretson
Adam Engebretson :
ASD-STE100 and Orwell's Rules?
2026-08-18 01:39:20
6
flyoverfarms
Great Value Rip Wheeler :
I did it for a day and I found it to be rude and terse😂 set back to default
2026-08-18 15:06:59
0
bellaciao6003
Bella Ciao 🇨🇦 :
or you could do like this dude on Reddit who made his Claude talk like a caveman to save tokens 😂
2026-08-18 02:50:33
8
kristakaista
Krista :
I’m going easier route. ”Speak like Loki.” or ”Speak like Geralt of Rivia.” 😅 And there is no dull day ever.
2026-08-18 04:51:24
8
samazure
Sam :
Actually, I love the fact that Claude pushes back most of the time. Sometimes it's right, sometimes it's wrong. However, this critical thinking aspect is amazing.
2026-08-18 00:23:07
0
strand917
strand917 :
I don’t think grok does that with me.
2026-08-18 08:13:58
0
kuhlsnu
kuhlsnu :
You should never change how the agent thinks and how it writes is also how it’s inference works. You can add that as a block at the end after it gets done with its inference or you’ll kill your benchmarks. I feel like you should know this.
2026-08-18 04:49:23
3
t2007v
T :
Okay question though…why a skill? This feels like a rule. Also thank you, I have to tell it to be concise like 2-3 times.
2026-08-18 12:27:33
4
ferdingler
ferdingler :
It doesn’t work reliably. I guarantee. You’ll have to remind it over and over to use it. It’s annoying.
2026-08-18 02:42:30
1
gumballjoe
Zero Cool aka Crash Override :
Great tip for documentation purposes, but AI models have a limited amount of processing power per response. If a huge portion of its brain is forced to constantly check, format, and filter its words to match a strict guide, it has less brainpower left to solve your actual problem
2026-08-18 11:13:11
4
larrygmaguire
Larry Maguire | AI Trainer :
it doesn’t work because the model is literally built to respond this way and skills are only loaded after the fact, and perhaps not at all in any given conversation. Claudish (apt word) is built into the system prompt so any results using skills are temporary and not robust I have found.
2026-08-18 00:22:57
2
9955rdkguh8643
23986430718634 :
Mine says “I’m going to have to push back on that” Bych just do what I tell you to do. I’m mad at him right now
2026-08-18 13:31:55
1
To see more videos from user @nate.b.jones, please go to the Tikwm homepage.

Other Videos


About