Sabitlenmiş Tweet
Quiet Take
I’m not a founder.
Not a VC.
Not here to shill you bags.
I’m just a guy who’s been deep in AI infrastructure, voice agents, and onchain UX for the last few months — actually using this stuff daily.
English
LifGenii Inc.
1.1K posts
LifGenii Inc.
@LIFgenii
AI Infrastructure & Onchain UX | Real agent adoption • Testnets & early mechanics
United States Katılım Mart 2016
806 Takip Edilen323 Takipçiler
@iruletheworldmo Benchmarks are finally starting to test whether agents can survive real work instead of just speedrunning toy tasks.
English
finally, a great benchmark. now we just need one thousand more.
seriously though, this is great.
and we do need one thousand more.
Serena Ge (Datacurve)@serenaa_ge
Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks. On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.
English
@milesdeutscher @base Crypto UX gets weird once agents can actually press buttons for you instead of just giving advice in chat.
English
Base ( @base ) just launched its new AI MCP - and it's extremely powerful.
AI agents can now trade on your behalf, interact on-chain, access agent skills & way more.
If you're in the crypto markets, you need to set this up.
🧵: How to set up the new Base MCP in <2 minutes.👇
English
@SoveyX Baseball easily has the best “just sitting there existing” atmosphere of any sport. Half the experience is the downtime.
English
@victormustar Tiny video models getting this usable this fast is kinda wild. A year ago this would’ve needed a massive stack just to demo.
English
cool new release: a tiny open video VLM that understands what happens in videos and when 👀
Marlin-2B (Apache 2.0!) can caption clips into timestamped events, or find a natural-language moment inside the video (can see a ton of cool use cases with it)
Made a Hugging Face demo for it ⬇️
English
@RoundtableSpace Agencies still billing like every line of code is handcrafted furniture.
English
A 28-YEAR-OLD IN BERLIN RUNS A 7-AGENT SOFTWARE FACTORY FROM HER PHONE AND UNDERCUT A $74K AGENCY QUOTE BY DELIVERING THE SAME WORK IN 19 HOURS FOR $28K.
The agency was still revising their proposal.
English
@fastlifejonny Doomscrolling evolved from a bad habit into full infrastructure.
English
@blknoiz06 Market absorbed a casual $1.3B like it was somebody testing a limit order UI.
English
@AvengedStoic @xai Running local subagents for the heavier tasks is probably gonna become way more common. Hybrid setups already feel surprisingly good.
English
@morganlinton @xai Fast feedback cycles are rare tbh. Most teams take weeks just to acknowledge obvious issues.
English
Loving Grok Build so much, amazing to see how much the team cares about feedback. In all honesty, never seen any company so actively involved in getting feedback, and acting on it so quickly.
Makes me really proud to be a Grok Build user since day one, I'm not going anywhere, and you can bet I'll keep sharing feedback like crazy!
English
@xai Usage limits are becoming the new benchmark. People actually want to live inside these tools now instead of just testing them once.
English
@auradeluxe The finale was good but they definitely let Homelander off too easy for someone that horrifying for 5 seasons.
English
Since we’re still talking about The Boys finale
I thought it was great.
If I were Kripke, and I’m not, the only two things I would have done differently would have been
1) make Homelander do one more truly grotesque and dastardly thing earlier in the episode just to remind us all of how truly awful he was and
2) make Homelander suffer more. Stretch out his death. Have Ryan take out a chunk, maybe MM and Starlight get a few blows in, too. It was just too quick and clean IMO
Otherwise, chef’s kiss. Great way to end it.
English
@auradeluxe LLMs struggling with letter counting while people are asking them to run entire companies now.
English
Oh… when this bubble pops it’s gonna pop hard
Naomi Rohatyn@omibaloney
Jesus. Fucking. Christ. On a bike. I just did this right now in 30 seconds. Just pick any word with 4 or more syllables.
English
@AmericaPartyX People still think this is some distant future thing while coding agents already replaced half my browser tabs.
English
🚨 ELON MUSK: AI AND ROBOTICS ARE A “SUPERSONIC TSUNAMI”
Elon Musk just warned that AI and robotics will bring the biggest transformation humanity has ever seen.
“This is really gonna be the most radical change that we’ve ever seen.”
He called it a “supersonic tsunami.”
The future is arriving faster than most people realize. x.com/cb_doge/status…
English
@HermesAgentTips Qwen quietly becoming part of every serious coding stack now. Happened way faster than I expected.
English
One of the most current Qwen beast models is now supported on Hermes Agent! Let’s go! Qwen 3.7 Max
Nous Research@NousResearch
Qwen 3.7 Max is now supported in Hermes Agent
English
@0xLuxi The only people consistently winning are the sportsbooks. That part never changes.
English
@kimmonismus People spent a year arguing benchmark scores while Codex users were just shipping stuff the whole time.
English
It's truly amazing to see how the general sentiment has shifted in favor of Codex.
I'm reading so many posts saying that Codex is really good now with GPT-5.5, and that Claude Code is regularly preferred.
(I've become a huge Codex fan myself).
At the same time, the new DeepSWE benchmark shows that GPT-5.5 is now ranked number one in this measurement as well.
Serena Ge (Datacurve)@serenaa_ge
Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks. On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.
English
@cascade_xyz $1B volume during private beta is actually crazy. Most products don’t even get real usage feedback at that stage.
English
Cascade has surpassed $1B in trading volume since the start of Private Beta.
Over the past couple months, we’ve worked closely with our early traders and testers to continuously refine Cascade.
We wouldn’t have reached this milestone without all of your constant support and feedback.
English
@cb_doge every tech company starts sounding a little less idealistic once the numbers get big enough. the AI race is basically geopolitics and trillion dollar incentives now, not just “build cool tools for humanity” anymore
English
NEWS: Sam Altman & OpenAI caught in self-dealing scandal.
60+ civic groups just demanded California’s AG investigate their shady Cerebras deal.
• Sam Altman & Greg Brockman are large personal investors in Cerebras
• OpenAI signed a multibillion-dollar deal to buy massive computing from them
• OpenAI’s spending exploded Cerebras valuation from ~$8B → $23B → ~$30B IPO
• Greg Brockman hid his personal stake from Elon Musk — under oath
• This is exactly the self-dealing California law (Corp Code § 5233) prohibits
OpenAI was founded as a nonprofit to benefit humanity. Its leaders now look like they’re directing company billions to pump their own investments.
The mission was always a lie.
English
@Teknium qwen updates inside coding stacks now feel like graphics card drivers for developers. install patch, suddenly your workflow moves 20% faster overnight
English
Got that latest Qwen for ya in Hermes Agent now!
Nous Research@NousResearch
Qwen 3.7 Max is now supported in Hermes Agent
English
One thing @NousResearch keeps getting right is model selection for actual agent workflows, not just leaderboard screenshots.
Seeing Qwen 3.7 Max added to Hermes Agent makes a lot of sense. @Alibaba_Qwen quietly built one of the strongest open models for long-running agent tasks right now — especially around tool usage, coding consistency, and sustained context handling over time.
35 hours autonomous runtime and 1k+ tool calls without collapsing is kinda the part people should pay attention to.
Nous Research@NousResearch
Qwen 3.7 Max is now supported in Hermes Agent
English