GCP: We don't need support, our product is too good
AWS: We don't need product, our support is too good
Azure: We don't need product or support, our sales is too good
3
51
459
We just published our paper on GPT-3! arxiv.org/abs/2005.14165
Proud to be part of this awesome team!
5
77
451
Proud to share what my team and I have been working on! I use Claude Code almost every day, and wouldn't want to code without it. Just getting started.
Claude Code has become indispensable for our team. In early testing, Claude completed tasks in a single pass that would normally take 45+ minutes of manual work.
Join the limited preview: docs.anthropic.com/en/docs/a…
ALT Claude Code welcome screen in a terminal window with stylized ASCII art logo.
5
5
133
11,552
I had a great chat with @lennysan ! We talked about everything from the future of AI and the potential for superintelligence by 2028, to the critical importance of AI safety and why we left OpenAI to start Anthropic. If you're curious about what a world with superintelligence could look like and how we're working to make it safe, give it a listen!
Listen to the full episode here: piped.video/WWoyWNhx2XU?si=WE1n…
11
7
129
19,153
Excited to announce what we've been working on the last few months: @AnthropicAI! We're looking for aligned researchers and engineers to help build reliable, steerable, powerful AI systems. Ping me if interested, or apply on our site.
2
10
85
This research has been a long time in the making. Super proud of the team! Understanding how language models work mechanistically is the first step to being able to modify that behavior at runtime at a semantic level. This will be a game changer in ensuring AGI is safe.
New Anthropic research paper: Scaling Monosemanticity.
The first ever detailed look inside a leading large language model.
Read the blog post here: anthropic.com/research/mappi…
ALT Title card for Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet
3
54
4,412
I've been using versions of Opus 4 for coding for a bit now. The numbers in the chart don't do it justice. Absolute game changer in staying on track, writing maintainable code, and generally doing what I want and expect rather than reward hacking. A+ job, team!
Introducing the next generation: Claude Opus 4 and Claude Sonnet 4.
Claude Opus 4 is our most powerful model yet, and the world’s best coding model.
Claude Sonnet 4 is a significant upgrade from its predecessor, delivering superior coding and reasoning.
ALT A benchmarking table titled Claude 4 benchmarks comparing performance metrics across various capabilities including coding, reasoning, tool use, multilingual Q&A, visual reasoning, and mathematics.
1
2
28
2,718
Love to see this!! Since we first prototyped Claude (and before it was named that), one of the most surprising and impressive things was how empathetic it seemed to be, and that's only improved over time. Try it and see ❤️
New Anthropic Research: How people use Claude for emotional support.
From millions of anonymized conversations, we studied how adults use AI for emotional and personal needs—from navigating loneliness and relationships to asking existential questions.
ALT Bar chart showing percentages of Claude conversations by type: Interpersonal advice (2.26%), Coaching (1.13%), Psychotherapy or counseling (0.34%), Companionship (0.31%), Romantic roleplay (0.05%), and Sexual roleplay (0.02%)
4
2
23
3,454
New episode drop: @8enmann, cofounder @AnthropicAI
- Claude 4 launch
- what happens when you act like tokens are free
- MCP
- specialization and the future of models
- safety for agents
- economic turing tests
1
21
8,394
This is the real game changer. Anyone can search the web, but searching your own data is a pain. Claude's really good at it.
Claude can also now connect with your Gmail, Google Calendar, and Docs.
It understands your context and can pull information from exactly where you need it.
ALT Claude interface showing Claude searching calendar and Gmail, with progress indicated by a loading spinner.
1
16
2,174
CRISPR, according to my friend who works on it at Synthego; commercial rocket flight; self driving cars' impact on real estate prices; many more
16
I've been using our internal prototypes of our MCP-enabled research workflow. It's amazing what Claude digs up. No more manual context stuffing! Hook up your data and give it a shot
Today we're announcing Integrations, a new way to connect your apps and tools to Claude.
We're also expanding Claude's Research capabilities with an advanced mode that searches the web, your Google Workspace, and now your Integrations too.
1
1
15
2,222
It's been great working with you @henrythe9ths ! Let's get some more founders in here 🚀
4
1
15
3,107
Which IDE do you use as your daily driver?
39%
VSCode
39%
Cursor
4%
Jetbrains
18%
Other
28 votes • Final results
1
7
447
I'm one of the first authors of GPT-3 and a founding member of Anthropic. You can definitely have impact now, DM me if you want to chat.
8
284
GPT-2 is why I went back to OpenAI after being laid off 2 years before. Then ended up building GPT-3 🎉
1
6
1,038
Pricing is the same as our normal models
cdn2.assets-servd.host/anthr…
5
111
I just published “Why AI research may be accelerating faster than experts realize” medium.com/p/why-ai-research…
3
I just published “How I made this impermanent wall art” medium.com/p/how-i-made-this…
3
Working on it!
In "Language Models (Mostly) Know What They Know", we show that language models can evaluate whether what they say is true, and predict ahead of time whether they'll be able to answer questions correctly. arxiv.org/abs/2207.05221
ALT A 52B language model can evaluate the validity of its own proposed answers - separating the correct and incorrect responses - on questions from TriviaQA, Lambada, Arithmetic, GSM8k, and Codex HumanEval. We have weighted the overall contribution from each of these five datasets equally.
1
3
I just published “The best dark chocolate 🍫” medium.com/p/the-best-dark-c…
3
All of section 4 and an additional appendix discuss this issue in the paper. Generally we found it to matter for a few datasets, which we marked with asterisks in our results section.
2
I just published “How to do everything wrong on your second solo backpacking trip” medium.com/p/how-to-do-every…
1
3
I just published “A game console for the internet” medium.com/p/a-game-console-…
1
2
I just published “How I lost 20 lbs in 3 months” medium.com/p/how-i-lost-20-l…
2
I just published “What Her teaches us about being human” medium.com/p/what-her-teache…
2
Thermometer -> temperature, cite amazon.com/Profiles-Future-I…
2
I just published “What I learned at a 10 day meditation retreat” medium.com/p/what-i-learned-…
2
I just published “The best chocolate ice cream brand” medium.com/p/the-best-chocol…
1
2
5 friends tried to draw a bike from memory after reading arxiv.org/abs/2006.06666
Mine is the top left.
2
I just published “9 questions to meditate on” medium.com/p/9-questions-to-…
1
2
- Anomaly detection (blood?)
- self cleaning or at least "clean me" alerts so it always gets a good image
- frequency would be a challenge if you don't always use that toilet. Some way to log manually
- real time courtesy flush rec 😉
1
57
Actually, success is everything other than growth. Growth is the necessary evil you tolerate in order to continue to scale with the needs of the business. For my teams in particular, maintaining and innovating on a service used by millions of people just isn't possible w 4 eng.
2
326
Actually, today's AI systems already consume human attention and interaction eg YouTube, FB, etc content recommendation systems
1
2
I just published “How to decide what your team should do” medium.com/p/how-to-decide-w…
2
I just published "How to run for people who hate running"
benjmann.net/how-to-run-for-…
1
I'm also doubtful of BCI/humans catching up. Timelines are too short, FDA too restrictive, wetlabs too hard.
If anyone uses recursive self improvement with unlimited capital then we're probably screwed. Otherwise relying on alignment techniques is my top choice
1
I just published “Who are your real friends?” medium.com/p/who-are-your-re…
1
IIUC, when using a good reward model, argmax actually works fine arxiv.org/abs/2009.01325
1
1
I just published “I researched model-based learning this summer. Here are the results.” medium.com/p/i-researched-mo…
1
1
I think her question was specifically about startups like Starcity, rather than houses that never meant to franchise
1
1
Prompt engineering is a hiccup in the development of good tech, unless you also consider the skill of asking a well meaning human for what you want prompt engineering.
1
Great feedback! Just getting started here. It does support multimodal input - you can either paste an image right into the TUI or you can reference an image path on disk.
1
317
I just published “What I learned in physical therapy” medium.com/p/what-i-learned-…
1
I agree that tweet is too glib, but even the other tweets in the series are more nuanced
This means that if we have a target behavior (e.g. non-discrimination) we may be able to nudge models to achieve that target using IF/CoT prompting if RLHF alone is not sufficient. But we must be careful to check whether RLHF + prompting causes the models to overshoot the target.
1
1
196
Cal Newport makes this argument well with more nuance in "So Good They Can't Ignore You", where instead he says "gain rare and valuable skills, and passion will follow"
1
1
I just published “When eating a burrito is a terrible idea” medium.com/p/when-eating-a-b…
1
Why burn to the ground when you can cut it down and sell the wood to be built into structures for extra sequestration?
1
1
I just published “How to do everything wrong on your first solo backpacking trip” medium.com/p/how-to-do-every…
1
You can apply the same argument to productivity/procrastination, hence complice.co/
1
I just published “I’m joining OpenAI to create human level artificial intelligence” medium.com/p/im-joining-open…
1