Julie Loves Tech

not an industry expert. just a dev and gamer who goes down rabbit holes and shares what she finds. 💻sometimes it's an AI tool i've been testing. 💻sometimes it's why a game shipped at 20fps and how a modder fixed it in a week. 💻sometimes it's a tool you can use at home that i wish i knew about sooner. no credentials. just curiosity and way too much screen time.

the best tech innovations don't just make things faster. they solve real problems, improve everyday life, and open doors to possibilities we haven't explored yet. 💻

technology never stands still. every day brings a new tool, idea, or breakthrough that changes how we work and live. the best thing you can do is stay curious, keep learning, and never stop adapting.

the AI race is expanding beyond software. 🤖 Foxconn is reportedly working on a new partnership with OpenAI while expanding humanoid robot deployments in its factories. the future of AI will be built through both intelligent software and advanced hardware.

watching hardware architecture and software compilation merge closer together shows that the next big performance leaps will come from co-designing chips and models from the ground up

openai claiming they have secret internal models that blow their public releases out of the water while simultaneously postponing their ipo to chase thirty billion in bridge financing feels like peak tech bubble theater

dropping error rates on t-states means quantum architectures won't have to waste up to 95% of their capacity just generating and purifying magic states anymore. massive leap toward practical scaling.

every new ai app says “your personal assistant” and then asks me to write the prompt. girl, i’m the assistant here

nothing like watching a $200B valuation frontier model spend 45 seconds overthinking a basic database query, only for the user to completely ignore its elaborate reasoning trace and type "nah make it shorter"

solana:8csyWk4JgpXACVKA3GzsQKksGahwme4MSmcNeFtCpump pushing aggressively into the traditional client computing space with its upcoming RTX Spark AI PC rollout is starting to rattle the long-standing duopoly held by intel and amd in consumer silicon.

microsoft just dropped a security patch fixing nearly a thousand vulnerabilities at once, proving that maintaining modern software infrastructure is basically a full-time emergency response.

just saw that uber cut 3,300 jobs specifically to fund robotaxis. not "ai will change the industry someday," they're doing it right now to their own headcount

so today my phone's battery health dropped 2% overnight and now it wants me to grieve

just found out a study finding significant gaps in human oversight of ai agent commands is the story that should worry people more than any benchmark score this week... crazyyyy

100% on ARC-AGI-3 with zero explicit instructions is wild, but the memory-across-attempts part is what actually gets me. every agent i've tested basically forgets its own failed attempts the second the context window fills up, so watching one actually build on what it tried before instead of starting fresh every time feels like the real unlock here. going to look into how AVO's memory setup actually works, that's the part worth understanding.

been getting hit with random network errors on ox alpha for weeks and just assumed it was a me problem. turns out 1.18.21 fixes it by auto-retrying unknown stops, small patch note, real difference if you've been rage-quitting mid-task like i have. updating right now before i lose another session to this.

if this gta 6 leak is real, it's the most significant gaming/fintech crossover since axie infinity. that's a big if. the claim: an in-game currency that can be traded for real-world money. no verified source, no rockstar confirmation, standard leak credibility disclaimer applies. but the concept itself is worth thinking through - because the infrastructure for this already exists and the regulatory environment is shifting fast enough that it's no longer impossible. what would actually need to be true for this to work: rockstar would need a money transmitter license or a fintech partner in every jurisdiction. that's not a small lift the currency would need anti-money laundering controls by law - which means kyc inside a video game the SEC and CFTC would have opinions about whether in-game tradeable assets are securities. they always do player-driven economies in games have failed before - diablo 3's real money auction house was shut down after 18 months the version that actually works: a stablecoin-based reward system where rockstar controls the conversion rate and can shut it off at any time. that's less exciting than the leak implies but more legally survivable. gta online already generates more revenue than most standalone games. add real money conversion and you change the entire incentive structure of how people play. waiting for rockstar to say something before forming a real opinion. but this is the leak worth watching if any of them are.

everyone talks about GPUs when they talk about the AI buildout bottleneck, but power delivery is quietly the thing actually slowing everything down. moving the electrical work into the factory so it ships pre-wired is such a boring-sounding fix for a genuinely huge problem. going down a rabbit hole on how much of the "AI is slowing down" narrative is actually just... infrastructure catching up.

the impressive part here is that the distance between “i have an idea for a game” and “my kid can actually play it” is collapsing to almost nothing. three.js used to require real development time; now the hard part is increasingly deciding what makes the game fun.

if this greyscale test is real, the timing is the interesting part.anthropic clearly isn't waiting around after fable 5. a successor being quietly tested on the existing fable 5 surface would suggest they're already validating the upgrade before flipping the switch broadly. the real question isn't whether 5.1 exists - it's whether the jump is big enough that users immediately notice.

openai just gave chatgpt eyes on your entire desktop. here's what that actually means for how you work. computer history is not the same as conversation memory. conversation memory is chatgpt remembering you prefer bullet points. computer history is chatgpt knowing you had figma open for three hours, switched to notion, then googled "design token best practices" before asking your question. that context difference changes the quality of what it can do. instead of you explaining your setup every session, it already knows your workflow. i've been testing ai tools long enough to know the gap between "feels more personalized" in an announcement and "actually more useful" in daily usage is usually significant. the personalization has to be accurate and relevant, not just present. what i'll be testing: whether it correctly infers context from app usage or just adds noise how it handles sensitive apps being open - health, banking, personal messages whether the "less explanation required" claim holds up on complex multi-tool workflows the opt-in framing is important but the default experience matters more. most people won't read the permission screen carefully.