English
Sergey @ science
831 posts
Sergey @ science
@sergey_science
Software engineer and founder with 25+ years of shipping systems end-to-end. Molecular biology, genetics, LLM Pipelines, Agents.
Finland انضم Şubat 2023
94 يتبع55 المتابعون
@x_ptlc @chrisramsay52 Oh stfu, you know its surprisingly hot 🫵🏻🧐
😂
English
Me when I meet a grey. 👽
This makes more sense if you’ve watched today’s video. But it works either way.
Credit: Highstrangeness115 on ig
English
@chrisramsay52 Chris! WTH! I had to do age verification to see THIS :D
English
@DrBeaVillarroel @Zigmanfreud Beatriz, I don't understand why do you need to convince some random person on X. :)
English
@Zigmanfreud Are you using AI, Mr Ziegler? Do you believe in the existence of artificial intelligence?
English
Let me save everyone a lot of time on the UFO/Space aliens issue…
It is mathematically/logically impossible for us to have been visited by outside life because we have investigated all the planets close enough for them to have even theoretically left for Earth AFTER we existed!
Rapid Response 47@RapidResponse47
.@POTUS: I recently directed @SecWar to begin releasing government files relating to UFOs and unexplained aerial phenomena. I am pleased to report this process is well underway. We've found many very interesting documents — and the first releases will begin very, very soon. 👽
English
I wonder why the feedback is so contradictory and inconsistent.
Personally, I've experienced both very good thinking of 4.7 even at low effort, and quite bizarre decisions of 4.7 in High effort mode.
Feels like there is dependency on time of the day even -maybe inference is controlled by the total load on the system and makes the model dumber "adaptively".
English
After 2 days of using Opus 4.7, I can say that this model feels like a step sideways, if not backwards. I've seen it:
- regress in its ability to write scientific papers
- hallucinate repeatedly
- fail to draw basic inferences about what I want
It feels different, not better
Jeremy Howard@jeremyphoward
Wow I can already say after just 5 hours using @AnthropicAI Opus 4.7 that this is the first model that "gets" what I'm doing when I'm working. It feels aligned with me in a way no previous model did. (4.6 actively worked against me. I hated it. So this is *very* exciting!)
English
@0xBoku @claudeai @AnthropicAI @felixrieseberg Hi Felix. There might be a bug in CC that makes CC to ignore cyber approve.
English
Got cyber approved for @claudeai doesn’t matter, still just shoots down my requests. How lame of @AnthropicAI to shoot down the adversary simulation side of security. Opus 4.6 works but it’s becoming trash and broke several projects this week. Without offensive security, cybersecurity would still be in the Stone Age
English
@SethSHowes How many read depths did you achieve?
but yes, retarded take and written by AI post. If you're not in bioinformatics, chances are, you did it wrong and came to convincing but wrong conclusions.
English
I’ve wanted to do this for a decade.
But I never did - I refuse to give any company my DNA.
It is me.
So this week I sequenced my genome entirely at home. Literally on my kitchen table.
I never exposed my DNA sequence to the internet. Not at any point.
I used a MinION to do the sequencing (it’s smaller + weighs less than an iPhone).
I used open-source DNA models for the analysis (Evo2 and AlphaGenome) running locally on a DGX Spark and Mac Studio.
I traced mechanisms behind my family’s multigenerational autoimmune conditions that no clinician has been able to understand.
When I set out to do this I didn’t know if it would actually work. It does.
Your genome is the most private data you will ever have. You probably shouldn’t let it leave your house.
Patrick Collison@patrickc
I'm lucky enough to have a great doctor and access to excellent Bay Area medical care. I've taken lots of standard screening tests over the years and have tried lots of "health tech" devices and tools. With all this said, by far the most useful preventative medical advice that I've ever received has come from unleashing coding agents on my genome, having them investigate my specific mutations, and having them recommend specific follow-on tests and treatments. Population averages are population averages, but we ourselves are not averages. For example, it turns out that I probably have a 30x(!) higher-than-average predisposition to melanoma. Fortunately, there are both specific supplements that help counteract the particular mutations I have, and of course I can significantly dial up my screening frequency. So, this is very useful to know. I don't know exactly how much the analysis cost, but probably less than $100. Sequencing my genome cost a few hundred dollars. (One often sees papers and articles claiming that models aren't very good at medical reasoning. These analyses are usually based on employing several-year-old models, which is a kind of ludicrous malpractice. It is true that you still have to carefully monitor the agents' reasoning, and they do on occasion jump to conclusions or skip steps, requiring some nudging and re-steering. But, overall, they are almost literally infinitely better for this kind of work than what one can otherwise obtain today.) There are still lots of questions about how this will diffuse and get adopted, but it seems very clear that medical practice is about to improve enormously. Exciting times!
English
@waejay_ @jesse_vermeulen Might be depth of emotional involvement (what's at stake) has impact too :)
English
@sergey_science @jesse_vermeulen i can’t imagine doing my manager’s job of context switching across 8 direct reports and 6 projects, so makes sense why eventually 5 different agents/sessions is hard ¯\_(ツ)_/¯ maybe it’s a learned skill though
English
I don't understand why you even started it.
1. Unnecessary costs for GH
2. Unnecessary tokens burned/electricity/fossil fuel
3. Unnecessary attention grab from all here
Is it supposed to be fun or what? Sounds like irresponsible thing to do.
If you have extra tokens that you don't know where to put, I will gladly accept them to build a genetic tool, for example.
English
With my codemaxxed project surpassing 353,000,000 lines of code (not a typo)
I actually got a @Github cease & desist 🪦
"We've noticed that the repository is growing fast while committing very frequently. This looks like some sort of automated activity that serves no purpose."
English
@waejay_ @jesse_vermeulen Ah, that’s why I’m so fried in the evening. I thought it’s 5 simultaneous sessions.
English
when you drift away to do something else, it leads to more context switch, which leads to more context switch, which makes it harder to stay in deep work mode, so i just started intentionally doing absolutely nothing. it gives my brain time to think alongside claude
English
Felix, I’m worried you got an ill culture of daily shipping. People are reporting lots of issues.
If I were a product manager in your team, I would pull emergency stop lever and focus on bug-free releases.
With such high pace you are heading towards low reputation. I’m honestly concerned.
English
We ship new little improvements every single day, but this one was requested so much that I'm tweeting about it: Skip all permissions for Claude Cowork.
Use with care, brought to you by @dreamofabear
English
@ClaudeDevs Assholes.
You disgust us.
I hope you get sued, you abuse your customers then call it a bug.
You full well know you're doing this on purpose.
You lie more than Claude does.
English
We fixed a bug where rate limits on Claude subscriptions weren't properly adjusted for long context requests in Opus 4.7.
We've reset 5-hour and weekly rate limits. Enjoy Opus 4.7!
English
@caleb_kinmon @bcherny @firstadopter But my setup is pristine : no mcp, almost no skills. It’s clean like after first install. Maybe that’s why.
English
@caleb_kinmon @bcherny @firstadopter I don’t have that poor experience.
Used 4.7 at low effort setting and shipped three changes in my work. Honestly, I saw no regression on 4.7 so far. But I did see them in 4.6 and 4.5.
English
@RayFernando1337 You probably need to go back to basics: threaten it, say you have no hands and blind, and promise billions in rewards if it thinks hard :)
English
Wait, what happened to the Extended Thinking toggle on Opus 4.7?
Opened Claude this morning and the toggle I use every day is gone. It's now "Adaptive thinking, thinks only when needed."
Dug into the docs and on 4.7, adaptive is the only mode. The model decides per-request if it wants to think or not. Where is the way to force it on?
What does this mean for a Max user like me who lives on the phone and web (not Claude Code or the API)?
On 4.6, Extended Thinking on meant every answer got the deep reasoning. I pay $200/month for Max and I kept it on for my workflows, projects, etc.
On 4.7, every request kind of feels like a slot machine. Did the model think about this one? Is my request worthy enough of more thinking? What if I tell it to think harder...ultrathink...mega ultra uber giga think?? I don't know. Pull the lever and just...hope?
When it does more thinking with 4.7 it is a nice experience and I love what the team built. Just wondering out loud if there's a way for $200 Max user to force thinking on every request. Happy to pay for it.
Anyone else notice this?
English
@sergey_science Where? I’ve been trying to look would love any help
English
We’ve heard your feedback and we’re working on making it easier to follow everything that’s happening with Claude Code.
First, we’re introducing @ClaudeDevs, the official channel to follow for all updates on Claude Code and the Claude platform.
ClaudeDevs@ClaudeDevs
For the developers building with Claude, a direct line from the team. Follow for changelogs, API releases, community updates, and deep dives.
English
@trq212 @ClaudeDevs Any chance you can help me with my subscription? I got charged but haven’t received it (submitted a ticket abt this too)
English
This can't be right. You can see on X people have been complaining ("Opus got nerfed").
I noticed intelligence of 4.6 has degraded to the level of Sonnet (I used Windsurf IDE with explicit Opus 4.6), so had to switch to Codex 5.3, then got back to Opus 4.5 which worked better than 4.6.
English
@koomai @HackingDave Can confirm swear chart is flat, which is why I’m asking for specific reports
English
Think about all the orgs using Claude right now that have no idea how bad it has become over the past 4 weeks ago.
No statement from Claude - but a total revert to where the model was a year ago - which in comparison to when 4.6 got released is effectively last years AI model.
The amount of bugs, security issues, and complete destruction of production applications is going to be felt for quite a long time due to this.
Claude: nothing to see here.
English
@yiannis__p @theo I think you are confusing Anthropic with OpenAI actually. It's the opposite.
English
@theo I feel like no matter how much better it might be than 5.4 I don’t really care cause Anthropic is too toxic to work with. Limits and costs aside, forcing me to use the worst harness destroys any potential model gains over 5.4
English
English
