is writing itself 🤯
Did it go sentient?!
Ofc no. It's AI to solve tickets. I give it tickets and it opens PRs with the solutions.
The % of its code that is written 'by itself' is going up fast.
Robots improving themselves. We're there! 🦾🤖🏁
@johnrushx
👀 don't think we understand yet how software development will look like in 10 years
my guess would be less positions, higher pay, more product-focused work
@levelsio
→ NVDA 📈
→ Apple intelligence 😴
→ Claude 3.5 Sonnet new best model
→ OpenAI to delay new voice mode release
→ DeepSeek-Coder-V2, new open source model, on level with best models at coding
→ Music labels suing Suno and Udio (may set a precedent cross industry)
→ Etched
@GergelyOrosz
almost anything that requires 1. mid level of creativity and/or 2. text processing, can be sped up having an llm around (no properly crafted products, literally just access to an llm). Think:
copy
content creation
litigation
legal research
data analysis & insights
anything
@twofifteenam
we're definitely getting faster python!
3.11 was pretty focused on performance enhancements and 3.13 is supposed to actually drop the GIL and present a just-in-time compiler (FAST STUFF)
@VicVijayakumar
@catalinmpit
if people already built it: it probably makes sense ✅
if somebody is already successful at selling it: the market exists and is ready ✅
if many are already successful at selling it and price war had started / is done: ok maybe a bit late
@teortaxesTex
they can create novel content
given examples can be extended, the logical patterns being apparently localized and utilized
local continuous weak generalization seems to happen, is what I mean
not a novel alien mind, but not just regurgitation
@emollick
I remember playing CTF with the original browsing gpt and it was actually quite ok at that
wouldn't even need a serious agentic setup if you can explain the vulnerability and provide good examples I'd guess
when tech is there to offer complete personal assistant ai, we won't have just one
don't know why we always imagined it like that, but the persona / personality could change on demand so we will most probably prefer different ones for different moments
believing ai can't surpass human intelligence in every metric is believing in the paranormal
we aren't that special, can't be
there must be better ways
A class perspective on the whole Twitter situation I'll really try to remember: tech workers were in such favorable position it was culturally forgotten we were still seen, at the end of the day, as mere costs.
Because what they see is that it's harder to get rich quick from it, and that's their only barometer of success. The whole idea that a platform could have social value far beyond the revenue it brings in literally does not compute.
the deeper you get into a system the more its apparent formality concedes space to something else
intuited patterns, dynamic structures, something in movement
@johnrushx
Thanks for the openness John 🙏
Allow failure, it's part of the process. You are doing amazing and inspiring many of us. Take care of yourself. Maybe time to sell some of the products or figure out how to get more, better tech help?
Claude asked on the effectivity of some random "improve sleeping" pill ad
made me think, when people get surrounded by llm-driven assistants, we could pretty much kill misleading marketing
critical we keep them as truthful and unbiased as possible
@DellAnnaLuca
@TheBrad_Hatter
I think this is the question. I struggle to imagine this output was unexpected by those pushing the tax increase. Less powerful lobbies, more governmental control on everything finance?
Jack Dorsey says people should have more children because more people and improving technology means we will solve our problems in increasingly creative ways
@gnopercept
it's called hyper-priming
the extra cannabinoids you get with pot affects semantic priming (ability to associate ideas), allowing for weaker associations to become more apparent
@jxmnop
also works for any llm driven app with no proper sanitation
worst case, a bad actor can figure out how a complete agentic path looks like
prompt injection really is the new sql injection
🙌 Just scheduled LinkedIn content for the next couple of months
First time EVER scheduling content.
Used
@buffer
, felt smooth.
👀 For those going solo, how often do you schedule content?
Which tool?
👀 Are you keeping an eye on analytics since day 1?
I plugged
@PlausibleHQ
as soon as the first page related to went public - it was super easy
Now I'm effectively generating an intuition for what works and what doesn't to get those first clicks
📊 What
what
@bryan_johnson
is doing really looks like a religion at this point lol
→ he plays a guide, prophet like figure
→ strict rituals (routines)
→ strict adherence to a lifestyle
→ promises of transcendence
→ spawning a moral framework
→ interest in recruiting others
→
@dejavucoder
agree, bothersome
moved to oai desktop app for this
i think in general oai is a bit further in making the setup feel like a workspace instead of a sequence of unconnected chats. these details should be fixed soon :/
@teortaxesTex
i'm not even convinced we have proved discrete global strong generalization is not there at all (actual generalization towards symbolic structures out of the training set)
@mar_kar_
👏 Awesome, interesting work!
three small things that come to mind after reading it:
- In asking for both the boolean response to the claim and its explanation in the same interaction, the accuracy of both tasks may suffer. A rerun asking just for the boolean could be worth it
103yo asked about AI video gen dangers:
"Literature did that (showcase a lifelike reality that did not exist). It was exciting and it gave meaning to us at the time. Why shouldn't this? [...] as long as it doesn't confuse the moment with reality"
based
@hermittoday
normally it's about self-narrative fulfillment, for both parts
but being conscious about that unlocks the other conversation meta, which actually is about sharing knowledge
@tollyferro
@yacineMTB
I'll be with the team that goes gil-less when it gets to stable and never look back
but yeah I feel you. thx for the efforts you are a py hero 🫡
@twofifteenam
totally
also i feel the gil discussion has overshadowed everything else when discussing python+speed, looking forward to see what comes next 👀
crazy how little hq content has spawned actually comparing results between claude 3.5 sonnet and gpt4-o
i still feel they are very close and anthropic's UI, slightly ahead, has been key for general sentiment to change like this