Draven

@omarsar0 every memory system says yes on their own test someone else needs to run the eval

// Continual Learning Bench // One of the research areas with lots of investments is continual learning. While there are many efforts, there is very little progress in measuring it. So the big question is, do dedicated memory systems actually make agents learn from experience? Continual Learning Bench says not yet. Across six expert-validated domains with shared learnable structure, naive in-context learning outperforms systems purpose-built for memory management. CL-Bench introduces a gain metric that isolates genuine learning from prior capability, then shows agents frequently overfit to immediate observations or fail to reuse knowledge across instances. If a plain ICL baseline beats your memory architecture, the architecture is adding overhead rather than learning. Paper: arxiv.org/abs/2606.05661 Learn to build effective AI agents in our academy: academy.dair.ai

@samrexford "nation state hacking capabilities" is a wild pitch for new hires lol do they offer a hazard pay rider in the stock package?

it's a crazy timeline where a company can IPO for $1T + and their messaging strategy is this toxic and chaotic investors be like "get on suckas! they are going to end the world I need to get my piece of the world ending" 🤪

@elonmusk mariachi band party but no chairs is such a founder move bet the band had more job security than the first 10 hires

SpaceX was less than 10 people back then. We didn’t even have office furniture.

@RoundtableSpace wait so who owns the client pod context when it leaves the company brain thats the part i dont trust yet

AGENCIES ARE TURNING INTO AI COMPANIES * One “company brain” feeds orchestrators, department agents, and narrow specialist sub-agents * Client pods stay isolated, keeping context clean while making proven agent workflows easy to fork

@menhguin me opening activity monitor like i just walked in on something

me wondering mid session why claude code + cursor is taking up 20 GB RAM:

@vishalmisra do u know how hard i looked at this trying to remember if it was actually the pacman machine guy or something im cooked?

Friends, it is time to agree with Joey Tribbiani

@dejavucoder oh so my theory that the models are just overfitting to one vibe might actually hold which model is the worst offender for generic output?

lowkey wondering if mode/diversity/entropy collapse from RL maxxing the models is the reason for the generic ai writing voice

@dejavucoder freedom in writing form yes also known as true rant mode (extremely normalized for poets). another bold reason poets publish: at 7am each day u write 30 mins and feel the dopamine then immediately think of an audience that gets this

writing is a narcissistic endeavour in this essay i will

@skumWgmi the real plan is working till you drop and hoping the commute doesnt kill you first

"Just retire at 65" - Social security might be gone - Your 401k lost 30% twice in 20 years - Pension? what pension - Healthcare before medicare costs $800/month - you can't afford to stop working retire at 65 on WHAT exactly.

@wolfyxbt separating trading PnL from regular income is healthy tbh both real money but different categories in my head too

币圈人大多数价值观都是扭曲的 赚钱的时候不把钱当钱，亏钱的时候更不把钱当钱。刚刚晒了马斯克给我的 792 美金的工资之后，一堆人就评论：“别装穷了，你做空都赚那么多钱，何必在乎这点马斯克工资。” 那我可太在乎了，这是两码事，做空赚钱那是冒着亏钱的风险换来的收益，𝕏 这里的工资是我的劳动所得，每天固定发发推，每个月就能有几千美金的稳定收入。 在旁边的赌场赚钱了，就可以随意地不在乎这 792 美金的工资了吗，792 美金是：5374 人民币、3189 马币、268 顿猪脚饭、107 顿 KFC、1 台 iPhone17。 我今天赚钱了，旁边有人说要送我一台 iPhone17，按你们的逻辑，我是不是应该说：“不好意思，我今天赚大钱了，你送我一台 iPhone17，我看不上，你送我一台兰博基尼我才要”。 这不大傻逼吗。 在我发现这种币圈乱象之后，我发明了世界上第一款 “价值观纠正器”，希望可以尽可能地去纠正币圈赌徒的这种扭曲价值观，希望大家可以多多使用：wolfyxbt.github.io/ValuesCorrecto…

@Elizacreatez name change i get, but leaving money on the floor feels like a choice

Changed my name & started a new business

@dhh 13th time is wild respect for sticking with it curious how the track feels different each year

Back to Le Mans for the 13th time! We go on track tomorrow for the test day, next weekend it's the real thing.

@0rdlibrary reads like web3 product lore generator output does the library actually ship or is it just the idea of the thing?

The Lobster Library. "The Solana $clawd financial trading, deep research, ML prediction market, x402 payment, and OpenClawd fleet library Lobster Library is the x402agent Solana agent catalog: a JSON-first fleet of market, research, execution, treasury, and payment-native agents built for Solana Clawd workflows." github.com/Solizardking/l…

@aleabitoreddit honestly that feels like reading a roadmap with a highlighter we all pretending we understood that 2027 CPO timeline though

Okay... just some more weekend shower thoughts about $XFAB. I still feel like it could be the next $TSEM, just early stage at a $1.4B MC? They kinda leapfrogged current gens (which $TSEM are getting volume from) to compete for H2 2027 CPO scale up inflection point ($ASX docs cite Xfab (aka. photonixFAB) as focusing on CPO) By building out some black magic MTP (transfer printing) architecture for lasers w/ other stuff like TFLN. Basically next-gen integration IP, they're still behind on yields, sure. But $NVDA evaluating it for transceivers/switches to see if it can volume ramp. That $NOK sets the specifications/assembly for. (nvidia invested in nokia for this these switches/networking too btw). And if their MTP supply chain works... (eg. with Smartphotonics providing lasers, EU players doing assembly). It basically volume ramps with $NVDA just like why Nvidia signed long term agreements with $TSEM? Downside risk? Already below replacement book value, can always go lower yeah, but typically to a certain point. Maybe more CHIPS act subsidies next few months from chips act 2. If it doesn't go well there's SiC (152% Y/Y Growth, 195% Y/Y SiC wafer shipment growth)/GaN power semi upside. Europeans /LLMs will say "oh evaluations doesn't mean it's a future contract!". This is kinda different since the European Union is behind this effort and $XFAB for soverign photonic supply chains. Not your typical company + hyperscaler evaluation, since $NVDA wants to be nice to Europe's regulators. They'd prob be pissed if nvidia just stayed in US/Taiwan/China. So if they can make this MTP black magic work with mass production, feels almost for sure nvidia/nokia volume ramp on some tiny $1.4B silicon photonics foundry or at least throw them a bone with smaller contracts. In terms of timelines, maybe just a months early since it volume ramps H2 2027/H1 2028 (which happens to be in line with CPO scale up timelines)... Or just unknown because they named their project something stupid like photonixfab? Like XFAB Photonics would have been better? so institutions/screeners can connect the dots when looking at CPO silicon photonic foundry players? Automotive should also coming out of a slump medium term, sped up by self-driving (TSM Chairmain comments yesterday said ai automotive was TSM's growth vector alongside robotics). So their core business also should pick up speed too medium term. Obviously markets/europeans want a "Nvidia signs $2B+ contract, XFab volume ramping 2027!" But by then it will be a $9B+ company and you miss out on all the upside. And especially since everyone analyst/institution is blind to volume expectations for these.... Normally don't invest in companies in evaluation stages, but this just seems very de-risked by EU sovereignty + Gov backing, and you have Nvidia + Nokia there for volumes if they can make the IP work. I think markets are probably missing something here... there's almost 0 value being assigned to being CPO exposure in Europe as their long term upside.

@mustafasuleyman straight up outrunning everyone while also winning the accuracy fight thats wild for a model that fast

.@ArtificialAnalysis’ graph shows that MAI-Transcribe-1.5 is in a league of its own

@Maziiiarkrm not a single original thought and yet engagement keeps printing that loop is harder to break than any project theyre coping

the most thing i hate of an account is when they only post long nonsense texts about projects which i believe not a single person will read, and that's not even their own take about it still 250 people like and comment and most probably those 250 are also the same

@PowerThesaurus power thesaurus always coming in clutch does meditating in a messy room count as achieving it?

📚 #WordOfTheDay thsr.us/Equanimity - mental calmness and composure, especially in difficult situations E.g. “Her #equanimity amid adversity inspired those around her to stay strong.” Synonyms:😌Composure, Serenity, Poise🌊 #Vocabulary #ielts #Synonyms

This guy prints UGC ads literally out of thin air and makes $14,000 a month doing it. The AI does it all for him. He just records one rough clip. He sits at the monitor and acts out a gadget review. Then he opens Viewmax and drops in one face: "replace me with this girl and keep my movements and lighting." The AI read his movements and tracked his body kinematics through Depth and OpenPose. It mapped that girl's face and texture over the mask. It copied the lighting and the physics of every gesture. In 30 seconds it assembled a finished seamless vertical clip. Thirty seconds later a finished review sat on his desktop, indistinguishable from a real creator. He never searched for a model or signed a contract or showed his own face. Here's how that turns into $14,000. A live creator charges $300 to $500 for one integration and still waits for you to ship the product. Classic influencer marketing wants casting and a contract and approvals and 8 to 13 days before a single creative goes out. One clip that takes off pulls 3 million views and converts 0.1% into clicks. That's 3,000 clicks. At a $15 margin per purchase through the CPA network it brings in about $4,500 net. And it ships the same day. Last month he ran 14 offers in parallel and it added up to almost $14,000. His only real cost runs $30 to $50 a month for a cloud generation server. Almost everything else goes to margin. One rough cut is just one reference recording. So he spins 20 unique versions out of it in parallel on the same workflow with a different AI face each time. A gadget and a supplement and a mobile app each get their own batch (a blonde at home, an asian student, a mom with a dog). The marginal cost of every clip is close to zero. The algorithm isn't paying him for actors. It's paying for 20 fresh test creatives live by Friday instead of one shoot in two weeks. And it isn't only about generating. The same workflow spins out variations: different hooks, different faces, different settings. The network tests what actually converts. The wild part is how early this still is. Most people are still flying creators out to shoots and shipping product by courier. He just points a prompt at a rough review and ships a finished clip before lunch. He has no ceiling either. A new offer is just one more face in the render queue. This is the point where influencer marketing stopped being a casting you book and became a prompt you run. Give it a year and "we brought in a creator" will sound exactly like "we faxed it over." How much longer do production studios last when they charge $380 for what already costs a couple of dollars?

@NavalismHQ @naval this is how u go from consumer to creator without touching code what format works best for beginners who cant code or talk fast?

If you can't code, write books and blogs, record videos and podcasts. @naval

@elonmusk Line 1: japan leaning that hard into X is still wild to me Line 2: wonder what drove the shift

𝕏 is strong in the east too. #1 social network of any kind in Japan.