GPT-6 Astra

“Was good last week, horrible with astra, completely burned thru entire week on 20x rn in 5 hours”

“I got burned myself. Once I had AI organize data, I repeatedly and emphatically constrained it, and in the end it mixed in a fabricated but seemingly reasonable number among several real data points. The reasoning process looked airtight. When my boss asked me to find the source, I couldn't. Later when I checked the original data, I realized that number simply didn't exist.”

“It feels more like a routine update, rather than the so-called AGI being advertised. Many outputs are slightly better, it is faster, and it sounds a little more natural.”

“Astra used the official website, indicated the exact parking to use, and corrected and updated the prices”

“Astra will chew up tokens”

“Quote from the thinking *summary*. And it left out the helmet :(”

“It's also really slow, feels slower than 5.6 sol max, probably because the scale went up.”

“AI overuses short sentences, sometimes intentionally keeps them terse, resulting in AI writing that often feels breathless, with chaotic rhythm.”

“The beginning is good, the first paragraph is good, the more you go on the worse it gets.”

“Short stories are nothing. Wait until it can write a novel of at least 100,000 characters and still sell it, then come back and talk about this.”

“I've used it, and the situation you mentioned does happen. But I found that it starts to get lazy and finds ways to not finish the work, and begins taking shortcuts. This is actually not what I wanted...”

“Astra seems to be really slow. Maybe it intends to read more context. But from my experience it is definitely slower than 5.6 sol when handling same tasks.”

“Overall it's probably smooth if you don't over-criticize and require every sentence to be meaningful.”

“I asked it to deploy gpt6 locally for me and it couldn't even do that. Trash.”

“It is (uses way less tokens)”

“Opus was more accurate in terms of for example this small building in the right down corner”

“After using GPT6 Astra for a whole morning, I feel it's still just a manual labor model, completely lacking the feeling when Fable5 first debuted”

“top of the line has never been an excuse for horribly inefficient”

“you can say "why is it not committed yet?" and it will give you an explanation and say it's actually ready to be committed. But it won't commit unless you explicitly say so. Well thank God, because that's the correct behavior”

“The actual improvement is not that big and then they say AGI is advancing vigorously through marketing, now they're also desensitizing these model releases, after all the fuss it's just changing the definition of AGI. Though it really is a decent model”

“Tried it initially, it's very good in coding, but it doesn't overwhelm Fable 5.1”

“Astra is blocking quite a lot of legitimate cybersecurity work for me right now — which is pretty much the exact reason I signed up for TAC in the first place”

“Its cache creation is quite expensive, and calling sub-agents consumes quickly too. I had everything set to max and xhigh for testing. If you didn't enable fast or similar and feel like it's consuming quickly, it's most likely a problem with your environment, because based on feedback from those around me and online, everyone feels it's consuming much slower than expected”

“The restrictions are going to be a huge problem as it was rare with SOL, but when it happened it was super annoying fixing it”

“It goes to GitHub to find ready-made solutions [crying laughing emoji].”

“After evolving a few more versions, it will cost more than hiring a human to write novels.”

“Astra is on another level 🤯”

“A model with incomplete multimodal capabilities achieving AGI? No way.”