Bon Cloudboy (@BonCloudboy) profile picture

dev | never financial advice | opinions are my own

Wien, Österreich
Joined May 2014
Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)Photo by Bon Cloudboy (@BonCloudboy)
5.4 releases and suddenly my old codex session fucks up 3 builds back to back. curious
2
43
Narrative reversal inc ? The take that you don't need juniors with AI was always kinda weird. Juniors can now do senior workloads. -> Junior agency has expanded, senior agency has decreased.
55
deepseek reasoning output is always so good. they seem like the only models that don't hyperfocus on the latest prompt.
108
hyping up models based on benchmarks before you actually try them out 😴
29
Feeling lazy 🥱 might let 5.2 prompt claude code today
32
Software will go into an era where we have *too much* software/too many features - and the skill is reduced to understanding pmf. Just as nowadays we suffer from *too much* information every day and profit from better algos.
27
Bon Cloudboy retweeted
Don’t confuse familiarity with understanding; both feel smooth.
49
120
4
801
56,128
i scrolled 50+ comments. all joking about the clip. saw only one that mentioned that there was no audio. all others are bots (?)
LeBron’s opening line being…”First of all, happy…happy International Women’s Day” cracks me up EVERY time 😂😂
267
we should just get over it. humans hallucinate all the time. whatever
how is the hallucination problem not solved yet? why are models still hallucinating and lying to me?
40
no counter argument presented btw
remember this absolute bullshit lmao
34
this is how LLMs work as well. the more you try to control them the worse the output is. imo
i think about this once a day
93
I believe we will have product-specific LLMs, not domain-specific ones. In addition to our current general models. E.g. OpenAI's Codex model.
1
27
this is basically why people that ask for bigger context windows just don't get it. in many cases LLM replies become unusable if you go beyond like 5 prompts. benchmarks being single turn is insufficient and completely ignores how people actually use LLMs.
1
94
do not understand how people liked that sora 2 trailer. they show some nice scenes but why start with a sam altman that does not look like him and also has a way different voice ?
193
openai favored bias on this app is so strong it is starting to sus me out. hope some of y'all get paid.
21
inb4 they put it in the system prompt and that's why it loses its mind on failure
Looks like Sergey's "threaten the model's life" trick got out of hand
30
Seems to be stating the obvious. Doubt a model shop will go hard into the "I don't know" direction. I would guess it is much more difficult to add this classification.
Everyone is talking about this new OpenAI paper. It's about why LLMs hallucinate. You might want to bookmark this one. Let's break down the technical details:
1
42
I wish they would stop with the hype. I get this is less than before but we are clearly stagnating at least for now. Time for businesses to catch up and to build actual products around this. The more you hype the harder the bubble will pop.
Anthropic CEO, Dario Amodei: If the current exponential growth in AI capabilities continues for just 1-3 more years, models will advance from being like a "smart undergraduate" to going "beyond the frontier of human knowledge"
1
33
so good
By popular request: you can now branch conversations in ChatGPT, letting you more easily explore different directions without losing your original thread. Available now to logged-in users on web.
29
imagine the current edition of hype posting about openai's new agi discovery is about a grok heavy copy paste
Elon: He accepted an offer at OpenAI and then uploaded our entire codebase Guess we will see GPT 5 heavy this winter!
1
3
364