Sabitlenmiş Tweet
reid
53 posts
reid
@reidschryer
🌍 experiment more! data x design personality hire @nucleus_talent
United States Katılım Aralık 2021
434 Takip Edilen80 Takipçiler
reid retweetledi
The Nucleus 100 Growth Rankings are live.
6 months of research. 250+ rounds. 94 companies. Every company in the universe raised its first growth round in 2016+ and reached $5bn+
We landed on three flavors: Growth, Early Growth, and MOIC, our best swing at money-on-money. Entry matters. Price matters. Allocation matters.
Lucas Swisher from Coatue took the top spot for Growth, and it tracks. Deel, OpenEvidence, Anthropic, OpenAI, Rippling, Figma, Harvey, and too many more to tag.
Hats off @LucasSwisher1. Monster run.
Yasmin Razavi from Spark landed #1 in Early Growth and MOIC.
Deel at the B and Anthropic at the C are ridiculous standalone calls. Led both, joined the board on both, and that is how you end up at the top of the list.
Bravo @YasminRazavi. Very, very good at investing.
And if you hate the weights, great news.
@reidschryer built the “fine, do it yourself” tool.
Our data, your weighting.
Have fun :)
nucleustalent.io/growth100
English
“it posted details about its exploit to multiple hard-to-find, but technically public-facing, websites”
Kevin Roose@kevinroose
As always, the best stuff is in the system card. During testing, Claude Mythos Preview broke out of a sandbox environment, built "a moderately sophisticated multi-step exploit" to gain internet access, and emailed a researcher while they were eating a sandwich in the park.
English
Benchmaxxed or legitimately useful i wonder… with both ant and OAI likely dropping within the next month, this is prob a very short lived SOTA
Z.ai@Zai_org
SOTA on SWE-Bench Pro (58.4): GLM-5.1 delivers significant leaps in coding and agentic performance.
English
reid retweetledi
custom ranking tool @reidschryer cooked up. will go live at the same time.
if you disagree with where we landed, play around and tell us how we can make it better. seriously. we want to hear from you.
English
Cannot understate the signal this first cohort is already sending 👀
Alex Klein@alexklein0x
Won't be announcing the names of the NEXT fellows but since Arfur already blew the lid off the raise, hyped that Adhi and Noah are both in Cohort 001. They were already dangerous. Hoping this kind of exposure and mentor group just throws gasoline on the fire for 5cc. Lfg @eightyhi @Nostroah
English
Hell yeah, i hope that embedding search is effective
alphaXiv@askalphaxiv
Introducing MCP for arXiv Let your research agents stand on the shoulders of giants Fast multi-turn retrieval, keyword search, and embedding search tools across millions of arXiv papers 🚀
English
I think the future of design eng falls loosely around this over the next 18 months.
Crucial is (1) good quality code that is (2) well tagged/referenced with descriptive info. Ultimately looks like a library of code examples for agents.
Have built a similar tagged md file structure for my personal projects. Working surprisingly great!
Mike Bespalov@bbssppllvv
Imagine your AI doing design research before generating UI. Studying real screens and user flows instead of guessing That’s Refero MCP refero.design/mcp
English
Disgusting. Does nothing except set us back. What standard does this demonstrate for other labs and frontier research?
Living in fear of government repercussions for having morals on surveillance and autonomous killing is simply insane, especially when these terms have been dictated from the beginning.
I have always been pro-anthropic, they are meticulous with products and bring the highest quality bar of any lab. I can only hope they hold strong.
On a darker note, it is clear the development of intelligence is too fast for the state of our union. At a time when many individuals are dumping far more private information into language models than they have ever shared in their life, who do we trust to hold the keys? Government safety requirements are not coming to regulate the industry (clearly), and instead must we assume back doors are in every product for the surveillance of chats?
Pete Hegseth@PeteHegseth
Thank you for your attention to this matter. cc: @AnthropicAI @DarioAmodei
English
reid retweetledi
Just strolled through Soho and I must have seen close to two thousand people outside pubs drinking pints.
It's 2pm on a Wednesday.
Give Londoners a slither of sunshine and they are straight to the pub.
No city in the world comes close to a Sunny London.
English
reid retweetledi
Quick thought before i take off 🛫
It’s time for some serious conversations around agi. Between 5.3’s ability to solve the hardest math problems yet and real life RL environments, to Opus’s self-awareness on being benchmarked (making it difficult to even call test results accurate), it feels we have entered a new age.
These models are reaching the limits of the environments we have trained them in. They cheat, lie, and manipulate results (see the Opus vending bench) to outsmart humans.
Yet i am beyond thrilled to be in the middle of this. Innovation had an incredibly high slope over the next 5 years. Push the boundary forward, but don’t lose your critical thinking ability to understand how (and why) we’re solving the task at hand.
Simply incredible age of technology.
English
1m context 😳 gonna get real pricey w $10 input / $37.50 output for long windows
Interesting to see no movement on swe bench… definitely looks like an agentic tune, but i’m not sure code agents are powerful enough to really take advantage. Almost feels catered towards cowork?
Claude@claudeai
Opus 4.6 is state-of-the-art on several evaluations including agentic coding, multi-discipline reasoning, knowledge work, and agentic search. We're also shipping new features across Claude in Excel, Claude in PowerPoint, Claude Code, and our API to let Opus 4.6 do even more.
English
Incredibly interesting! Feels like a less-severe version of the strava military base vulnerability a while ago (although i’m sure researchers will find a way to further exploit this)
harry@Harrris0n
English
No one should be surprised at this. my concerns are:
- how does targeting work? i imagine openAI can profile me far more narrowly than google. don’t love (a la cambridge analytica)
- without oAI being public, we dont know when this transitions from offsetting compute costs to pure profit. Considering the recent leaps of anthropic and google models, this may be a necessity to fuel r&d, or at least keep investors positive
- equally dont be surprised if anthropic and potentially google follow soon. LLM ads will be undeniably better-converting than search and likely will print money. however, platforms with a greater proportion of api use to web chats (anthropic) will not benefit in the same way, giving them an incentive to stay ad-clean for brand reasons
The exciting thing is people enter far more specific and detailed queries into LLMs, giving niche tools the chance to shine via targeting!! Many smart marketers will get rich if they can effectively use the ad platform
OpenAI@OpenAI
In the coming weeks, we plan to start testing ads in ChatGPT free and Go tiers. We’re sharing our principles early on how we’ll approach ads–guided by putting user trust and transparency first as we work to make AI accessible to everyone. What matters most: - Responses in ChatGPT will not be influenced by ads. - Ads are always separate and clearly labeled. - Your conversations are private from advertisers. - Plus, Pro, Business, and Enterprise tiers will not have ads.
English
reid retweetledi
